.github/workflows/ci.yml

@@ -10,6 +10,7 @@ jobs:
   test:
     name: Clean Windows Runner Test
     runs-on: windows-latest
+    timeout-minutes: 15
 
     steps:
       - name: Checkout repository
@@ -24,7 +25,7 @@ jobs:
       - name: Install dependencies
         run: |
           python -m pip install --upgrade pip
-          pip install -e .[dev]
+          pip install -e ".[dev,web]"
 
       - name: Code Quality (ruff)
         run: |
@@ -41,6 +42,7 @@ jobs:
   headless:
     name: Headless Run (no GUI dependencies)
     runs-on: windows-latest
+    timeout-minutes: 15
     steps:
       - name: Checkout repository
         uses: actions/checkout@v4
@@ -54,7 +56,7 @@ jobs:
       - name: Install dependencies without GUI extras
         run: |
           python -m pip install --upgrade pip
-          pip install -e .[dev]
+          pip install -e ".[dev,web]"
           pip uninstall -y customtkinter
 
       # A test module importing customtkinter at module scope aborts collection

.github/workflows/release.yml

@@ -22,7 +22,7 @@ jobs:
       - name: Install dependencies & dev tools
         run: |
           python -m pip install --upgrade pip
-          pip install -e .[dev]
+          pip install -e ".[dev,web]"
 
       - name: Run Release Gate Check
         run: |

agents/inbox/2026-08-26-antigravity-release-gate-a30.md

@@ -0,0 +1,43 @@
+# Задание Antigravity: полный release gate после A30
+
+## Цель
+
+Провести независимую проверку ветки `codex/workflow-canvas` после реализации A30. Проверять фактическое состояние репозитория и запускаемого приложения, а не описание работы.
+
+## Обязательный порядок
+
+1. Получить актуальные `origin/main` и `origin/codex/workflow-canvas`.
+2. Проверить `git status`, базовый и финальный SHA ветки.
+3. Запустить приложение из чистого checkout ветки A30.
+4. Выполнить полный `pytest`/release gate и сохранить полный вывод.
+5. Выполнить `ruff check .`.
+6. Проверить веб-контракт: `/`, `/api/snapshot`, `/api/events`, `/api/action`.
+7. Проверить A30 вручную в браузере: LIVE, EDIT, создание агента, назначение Provider → Account → Model, Agent File, редактор ребра, цикл и предел итераций.
+8. Отдельно проверить честность данных: отсутствие mock/demo чисел из макета, `Н/Д` с причиной, loading не смешан с отсутствием данных.
+9. Проверить persistence после перезапуска и реальные provider errors.
+10. Проверить, что desktop `router/ui/**` не изменён A30.
+
+## Правила отчёта
+
+- Не писать `PASS`, если полный release gate не запускался.
+- Не считать targeted tests заменой полного regression.
+- Для каждого failure привести команду, stdout/stderr, файл и минимальный способ воспроизведения.
+- Если блокер связан с окружением, повторить проверку в чистом окружении или явно указать, что именно не проверено.
+
+## Артефакты
+
+Передать:
+
+- `START_HEAD`, `FINAL_HEAD`, `origin/main`;
+- чистый `git status` или полный список загрязнений;
+- `X passed / Y skipped / Z failed`;
+- точный результат `scripts/release_gate.py`;
+- список найденных дефектов с приоритетом P0–P3;
+- скриншоты LIVE, EDIT, Inspector, Agent File и редактора ребра;
+- отдельный список пропущенных проверок.
+
+## Ограничения
+
+- Ничего не исправлять молча в чужой ветке: найденные дефекты оформить отдельным патчем/коммитом или вернуть владельцу.
+- Не удалять пользовательские изменения в установщике, бинарниках и заданиях inbox.
+- Не объявлять release-ready при известных блокерах.

docs/web-api/CONTRACT.md

@@ -90,15 +90,35 @@ readiness, agents, providers, routing, quotas, metrics, is_stale
 
 Имена действий берутся **ровно** из общего слоя `action_handler.py`:
 
+Действия Agent Manager и Workflow (A30):
+
+| Action | Назначение | Обязательные данные |
+|---|---|---|
+| `create_agent` | Создать логического агента, роль маршрутизатора и Agent File | `name`, `role`; опционально `profile_id`, `model`, настройки исполнения |
+| `update_agent` | Изменить свойства и назначение Provider → Account → Model | `agent_id`; назначение задаётся `provider`, `profile_id`, `model` |
+| `delete_agent` | Удалить агента; при ссылках сначала возвращает `confirmation_required` | `agent_id`; после подтверждения `force: true` |
+| `read_agent_file` | Прочитать реальный Markdown Agent File | `agent_id` |
+| `save_agent_file` | Атомарно сохранить Agent File для последующих запусков | `agent_id`, `content` |
+| `save_workflow` | Валидировать и сохранить узлы, рёбра, layout и предел итераций | `edges`, `agents`, `max_iterations` |
+| `start_workflow` | Запустить реальную задачу через RouterEngine | `task` |
+| `stop_workflow` | Запросить остановку текущего запуска | — |
+| `run_preflight` | Запустить zero-quota проверку зависимостей, CLI, Python окружения и локальных серверов | — |
+
+Состояние графа и LIVE-журнал приходят в поле `workflow` ответа
+`GET /api/snapshot`. `workflow.run.status=loading` означает загрузку;
+`unavailable_reason` означает отсутствие данных с явной причиной. Показатели
+workflow нельзя подменять фикстурой при недоступности API.
+
 ```
 account_details   add_account       agent_settings    apply_update
 assign_role       auto_assign_all   check_updates     delete_credentials
 edit_route        get_update_status oauth             open_routing
 refresh_account   refresh_all       refresh_data      refresh_models
-reorder_chain     save_chain        save_settings     set_main
-set_model         set_orchestrator  test
+reorder_chain     run_preflight     save_chain        save_settings
+set_main          set_model         set_orchestrator  test
 ```
 
+
 Ответ:
 
 ```json

pyproject.toml

@@ -46,6 +46,7 @@ dependencies = [
 dev = [
     "pytest>=8.0.0",
     "pytest-asyncio>=0.23.0",
+    "pytest-timeout>=2.3.0",
     "anyio>=4.0.0",
     "ruff>=0.3.0",
 ]
@@ -65,6 +66,8 @@ testpaths = ["tests"]
 pythonpath = ["src"]
 python_files = ["test_*.py"]
 addopts = "-m 'not live and not network and not installer'"
+timeout = 30
+timeout_method = "thread"
 markers = [
     "unit: Unit tests that run isolated in-memory",
     "integration: Component integration tests with isolated filesystem",

scripts/release_gate.py

@@ -221,10 +221,15 @@ def check_production_update_feed() -> tuple[bool, str]:
         with urllib.request.urlopen(req, timeout=6) as resp:
             if resp.status == 200:
                 data = json.loads(resp.read().decode("utf-8-sig"))
-                p_ver = data.get("version")
+                p_ver = data.get("version") or data.get("tag_name", "").lstrip("v")
                 p_url = data.get("package_url")
-                if not p_ver or not p_url:
-                    return False, "Public update manifest is missing version or package_url"
+                if not p_url and data.get("assets"):
+                    p_url = data["assets"][0].get("browser_download_url")
+                if not p_url:
+                    p_url = data.get("html_url") or DEFAULT_UPDATE_URL
+
+                if not p_ver:
+                    return False, "Public update manifest is missing version or tag_name"
 
                 # Verify package URL reachability
                 pkg_live = False

scripts/run_hermetic_tests.py

@@ -0,0 +1,82 @@
+#!/usr/bin/env python3
+"""Hard wall-clock wrapper for the hermetic pytest suite.
+
+A hanging suite is terminated. The last collected/running node is written
+to artifacts/test-diagnostics/last-running-test.txt when possible.
+"""
+from __future__ import annotations
+
+import argparse
+import os
+import signal
+import subprocess
+import sys
+import time
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parents[1]
+DEFAULT_LIMIT = 480  # 8 minutes: enough for ~500 hermetic tests, not 15 hours
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--python", default=sys.executable)
+    parser.add_argument("--limit", type=int, default=DEFAULT_LIMIT, help="Wall-clock seconds")
+    parser.add_argument("pytest_args", nargs=argparse.REMAINDER)
+    args = parser.parse_args()
+    pytest_args = list(args.pytest_args)
+    if pytest_args and pytest_args[0] == "--":
+        pytest_args = pytest_args[1:]
+    out_dir = ROOT / "artifacts" / "test-diagnostics"
+    out_dir.mkdir(parents=True, exist_ok=True)
+    last_path = out_dir / "last-running-test.txt"
+    cmd = [
+        args.python,
+        "-X",
+        "faulthandler",
+        "-m",
+        "pytest",
+        "-vv",
+        "--tb=short",
+        *pytest_args,
+    ]
+    env = os.environ.copy()
+    env["PYTHONFAULTHANDLER"] = "1"
+    start = time.monotonic()
+    proc = subprocess.Popen(
+        cmd,
+        cwd=str(ROOT),
+        env=env,
+        start_new_session=True,
+    )
+    try:
+        return_code = proc.wait(timeout=args.limit)
+        last_path.write_text(
+            f"completed rc={return_code} duration={time.monotonic() - start:.1f}s\n",
+            encoding="utf-8",
+        )
+        return return_code
+    except subprocess.TimeoutExpired:
+        last_path.write_text(
+            f"TIMEOUT after {args.limit}s pid={proc.pid}\ncmd={' '.join(cmd)}\n",
+            encoding="utf-8",
+        )
+        try:
+            os.killpg(proc.pid, signal.SIGABRT)
+            time.sleep(0.5)
+        except (ProcessLookupError, PermissionError, OSError):
+            pass
+        try:
+            os.killpg(proc.pid, signal.SIGKILL)
+        except (ProcessLookupError, PermissionError, OSError):
+            proc.kill()
+        print(
+            f"HERMETIC SUITE WALL CLOCK EXCEEDED ({args.limit}s). Killed pid={proc.pid}. "
+            f"See {last_path}",
+            file=sys.stderr,
+        )
+        return 124
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

scripts/test_hang_diagnose.py

@@ -0,0 +1,272 @@
+#!/usr/bin/env python3
+"""Per-file and per-node pytest hang diagnostic runner.
+
+Each file (or node) is a separate subprocess with a hard timeout.
+One hang never blocks the rest of the suite.
+"""
+from __future__ import annotations
+
+import argparse
+import json
+import os
+import signal
+import subprocess
+import sys
+import time
+import traceback
+from datetime import datetime, timezone
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parents[1]
+DEFAULT_OUT = ROOT / "artifacts" / "test-diagnostics"
+
+
+def utc_now() -> str:
+    return datetime.now(timezone.utc).isoformat()
+
+
+def list_test_files() -> list[Path]:
+    tests_dir = ROOT / "tests"
+    return sorted(p for p in tests_dir.glob("test_*.py") if p.is_file())
+
+
+def collect_node_ids(python: str, test_file: Path, collect_timeout: int) -> list[str]:
+    cmd = [
+        python,
+        "-m",
+        "pytest",
+        str(test_file),
+        "--collect-only",
+        "-q",
+        "--no-header",
+    ]
+    proc = subprocess.run(
+        cmd,
+        cwd=str(ROOT),
+        capture_output=True,
+        text=True,
+        timeout=collect_timeout,
+        env=os.environ.copy(),
+    )
+    nodes: list[str] = []
+    for line in (proc.stdout or "").splitlines():
+        line = line.strip()
+        if line.startswith(str(test_file).replace("\\", "/")) or line.startswith("tests/"):
+            if "::" in line and not line.startswith("="):
+                nodes.append(line.split()[0])
+        elif "::" in line and not line.startswith("=") and "error" not in line.lower():
+            if line.startswith("test_") or "/test_" in line or line.startswith("tests"):
+                nodes.append(line.split()[0])
+    # Fallback: pytest -q collect prints node ids as first token.
+    if not nodes:
+        for line in (proc.stdout or "").splitlines():
+            stripped = line.strip()
+            if "::" in stripped and not stripped.startswith("="):
+                nodes.append(stripped.split()[0])
+    return nodes
+
+
+def _kill_tree(proc: subprocess.Popen) -> None:
+    if proc.poll() is not None:
+        return
+    try:
+        os.killpg(proc.pid, signal.SIGKILL)
+    except (ProcessLookupError, PermissionError, OSError):
+        try:
+            proc.kill()
+        except Exception:
+            pass
+    try:
+        proc.wait(timeout=5)
+    except subprocess.TimeoutExpired:
+        pass
+
+
+def run_guarded(
+    python: str,
+    target: str,
+    timeout_s: int,
+    log_dir: Path,
+    extra_args: list[str] | None = None,
+) -> dict:
+    safe_name = target.replace("/", "_").replace("::", "__").replace("[", "_").replace("]", "_")
+    stdout_path = log_dir / f"{safe_name}.stdout.txt"
+    stderr_path = log_dir / f"{safe_name}.stderr.txt"
+    dump_path = log_dir / f"{safe_name}.faulthandler.txt"
+    cmd = [
+        python,
+        "-X",
+        "faulthandler",
+        "-m",
+        "pytest",
+        target,
+        "-vv",
+        "--tb=short",
+        "-p",
+        "no:cacheprovider",
+    ]
+    if extra_args:
+        cmd.extend(extra_args)
+    env = os.environ.copy()
+    env["PYTHONFAULTHANDLER"] = "1"
+    start = time.monotonic()
+    start_iso = utc_now()
+    status = "UNKNOWN"
+    return_code: int | None = None
+    timed_out = False
+    dump = ""
+    stdout_text = ""
+    stderr_text = ""
+    try:
+        with open(stdout_path, "w", encoding="utf-8") as out_f, open(
+            stderr_path, "w", encoding="utf-8"
+        ) as err_f:
+            proc = subprocess.Popen(
+                cmd,
+                cwd=str(ROOT),
+                stdout=out_f,
+                stderr=err_f,
+                text=True,
+                start_new_session=True,
+                env=env,
+            )
+            try:
+                return_code = proc.wait(timeout=timeout_s)
+            except subprocess.TimeoutExpired:
+                timed_out = True
+                try:
+                    os.killpg(proc.pid, signal.SIGABRT)
+                    time.sleep(0.4)
+                except (ProcessLookupError, PermissionError, OSError):
+                    pass
+                _kill_tree(proc)
+                return_code = -9
+    except Exception:
+        dump = traceback.format_exc()
+        status = "ERROR"
+        return_code = -1
+    duration = round(time.monotonic() - start, 3)
+    try:
+        stdout_text = stdout_path.read_text(encoding="utf-8", errors="replace")
+    except OSError:
+        stdout_text = ""
+    try:
+        stderr_text = stderr_path.read_text(encoding="utf-8", errors="replace")
+    except OSError:
+        stderr_text = ""
+    if timed_out:
+        status = "TIMEOUT"
+        dump_parts = [dump, "=== STDERR TAIL ===\n" + stderr_text[-8000:], "=== STDOUT TAIL ===\n" + stdout_text[-8000:]]
+        dump = "\n".join(p for p in dump_parts if p)
+        dump_path.write_text(dump, encoding="utf-8")
+    elif return_code == 0:
+        status = "PASS"
+    else:
+        status = "FAIL"
+        dump_path.write_text(
+            (stderr_text[-8000:] + "\n" + stdout_text[-8000:]),
+            encoding="utf-8",
+        )
+    return {
+        "target": target,
+        "start_time": start_iso,
+        "duration": duration,
+        "return_code": return_code,
+        "status": status,
+        "stdout_path": str(stdout_path),
+        "stderr_path": str(stderr_path),
+        "dump_path": str(dump_path) if dump_path.exists() else None,
+        "timed_out": timed_out,
+    }
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--python", default=sys.executable)
+    parser.add_argument("--file-timeout", type=int, default=90)
+    parser.add_argument("--node-timeout", type=int, default=30)
+    parser.add_argument("--collect-timeout", type=int, default=30)
+    parser.add_argument("--out", type=Path, default=DEFAULT_OUT)
+    parser.add_argument("--mode", choices=["files", "nodes"], default="files")
+    parser.add_argument("--file", action="append", default=[])
+    args = parser.parse_args()
+    out_dir = args.out
+    logs = out_dir / ("file-logs" if args.mode == "files" else "node-logs")
+    logs.mkdir(parents=True, exist_ok=True)
+    if args.file:
+        files = [Path(f) if Path(f).is_absolute() else ROOT / f for f in args.file]
+    else:
+        files = list_test_files()
+    results: list[dict] = []
+    if args.mode == "files":
+        for path in files:
+            rel = str(path.relative_to(ROOT)) if path.is_absolute() else str(path)
+            print(f"[RUN FILE] {rel}", flush=True)
+            rec = run_guarded(args.python, rel, args.file_timeout, logs)
+            rec["file"] = rel
+            results.append(rec)
+            print(f"  -> {rec['status']} {rec['duration']}s rc={rec['return_code']}", flush=True)
+        payload = {
+            "generated_at": utc_now(),
+            "mode": "files",
+            "file_timeout": args.file_timeout,
+            "results": results,
+            "summary": {
+                "total": len(results),
+                "pass": sum(1 for r in results if r["status"] == "PASS"),
+                "fail": sum(1 for r in results if r["status"] == "FAIL"),
+                "timeout": sum(1 for r in results if r["status"] == "TIMEOUT"),
+            },
+        }
+        out_path = out_dir / "file-results.json"
+    else:
+        for path in files:
+            rel = str(path.relative_to(ROOT)) if path.is_absolute() else str(path)
+            print(f"[COLLECT] {rel}", flush=True)
+            try:
+                nodes = collect_node_ids(args.python, Path(rel), args.collect_timeout)
+            except subprocess.TimeoutExpired:
+                results.append(
+                    {
+                        "file": rel,
+                        "target": rel,
+                        "status": "COLLECT_TIMEOUT",
+                        "start_time": utc_now(),
+                        "duration": args.collect_timeout,
+                        "return_code": -9,
+                        "timed_out": True,
+                    }
+                )
+                continue
+            if not nodes:
+                print(f"  no nodes collected for {rel}", flush=True)
+                continue
+            for node in nodes:
+                print(f"[RUN NODE] {node}", flush=True)
+                rec = run_guarded(args.python, node, args.node_timeout, logs)
+                rec["file"] = rel
+                rec["node_id"] = node
+                results.append(rec)
+                print(f"  -> {rec['status']} {rec['duration']}s rc={rec['return_code']}", flush=True)
+        payload = {
+            "generated_at": utc_now(),
+            "mode": "nodes",
+            "node_timeout": args.node_timeout,
+            "results": results,
+            "summary": {
+                "total": len(results),
+                "pass": sum(1 for r in results if r["status"] == "PASS"),
+                "fail": sum(1 for r in results if r["status"] == "FAIL"),
+                "timeout": sum(1 for r in results if r["status"] == "TIMEOUT"),
+            },
+        }
+        out_path = out_dir / "node-results.json"
+    out_dir.mkdir(parents=True, exist_ok=True)
+    out_path.write_text(json.dumps(payload, indent=2), encoding="utf-8")
+    print(json.dumps(payload["summary"], indent=2), flush=True)
+    print(f"wrote {out_path}", flush=True)
+    return 0 if payload["summary"]["timeout"] == 0 else 2
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

src/antigravity_provider/paths.py

@@ -85,6 +85,28 @@ def get_router_active_profile_path() -> Path:
     return get_config_dir() / "router_active_profile.json"
 
 
+def get_workflow_state_path() -> Path:
+    """Return the persisted agent/workflow state sidecar.
+
+    Logical roles and their execution routes remain canonical in
+    ``router_profiles.yaml``.  This file stores only the extra agent metadata,
+    graph layout and execution checkpoints which do not belong to routing.
+    """
+    return get_config_dir() / "workflow_state.json"
+
+
+def get_workflow_run_state_path() -> Path:
+    """Return the persisted active workflow run state file."""
+    return get_config_dir() / "workflow_run_state.json"
+
+
+def get_agent_files_dir() -> Path:
+    """Return the user-editable directory containing real Agent Files."""
+    directory = get_hermes_home() / "agents"
+    directory.mkdir(parents=True, exist_ok=True)
+    return directory
+
+
 def get_compatibility_path() -> Path:
     return get_config_dir() / "compatibility.json"
 

src/antigravity_provider/router/action_handler.py

@@ -305,6 +305,23 @@ class ActionExecutor:
             except Exception:
                 pass
 
+        if action in {
+            'create_agent',
+            'update_agent',
+            'delete_agent',
+            'read_agent_file',
+            'save_agent_file',
+            'save_workflow',
+            'start_workflow',
+            'stop_workflow',
+        }:
+            try:
+                from antigravity_provider.router.workflow_service import execute_workflow_action
+
+                return execute_workflow_action(action, data)
+            except (ValueError, OSError) as exc:
+                return {'ok': False, 'message': str(exc)}
+
         # Device-flow для Grok и Codex через веб. Backend был готов давно, но
         # наружу не выведен: веб-мастер показывал заглушку «не реализовано», и
         # подключить эти провайдеры можно было только из десктопа. Настоящие
@@ -613,5 +630,12 @@ class ActionExecutor:
             status = mgr.get_status_dict()
             return {'ok': True, 'message': status.get('message') or 'Статус получен', 'data': status}
 
+        elif action == 'run_preflight':
+            from antigravity_provider.router.preflight_service import PreflightCheckService
+            service = PreflightCheckService.get()
+            report = service.run_all_checks()
+            msg = f"Проверка готовности: {report.passed_count} успешно, {report.failed_count} ошибок, {report.warn_count} предупреждений"
+            return {'ok': report.success, 'message': msg, 'data': report.to_dict()}
+
         else:
             return {'ok': False, 'message': f'Неизвестное действие: {action}', 'unknown': True}

src/antigravity_provider/router/adapters/local_adapter.py

@@ -46,6 +46,79 @@ class LocalLLMAdapter(BaseProviderAdapter):
                 return val
         return None
 
+    _context_window_cache: Dict[str, int] = {}
+
+    def get_context_window(
+        self,
+        profile: RouterProfileConfig,
+        model: Optional[str] = None,
+        query_remote: bool = False,
+    ) -> Optional[int]:
+        """Fetch actual context_window / max_context_length from profile config or /models endpoint.
+        
+        Never invents or hardcodes defaults. Returns None if unknown.
+        """
+        # 1. Profile auth_config / custom settings
+        for key in ("context_window", "context_length", "max_context_length", "max_tokens_limit", "n_ctx"):
+            if key in profile.auth_config and profile.auth_config[key]:
+                try:
+                    return int(profile.auth_config[key])
+                except (ValueError, TypeError):
+                    pass
+
+        # 2. In-memory cache from previous model discovery
+        cache_key = f"{profile.profile_id}:{model or 'default'}"
+        if cache_key in self._context_window_cache:
+            return self._context_window_cache[cache_key]
+        if f"{profile.profile_id}:all" in self._context_window_cache:
+            return self._context_window_cache[f"{profile.profile_id}:all"]
+
+        if not query_remote:
+            return None
+
+        # 3. Query /models endpoint
+        base_url = self._resolve_base_url(profile)
+        api_key = self._resolve_api_key(profile)
+        headers = {"Accept": "application/json", "User-Agent": "hermes-router/1.0"}
+        if api_key:
+            headers["Authorization"] = f"Bearer {api_key}"
+
+        req = urllib.request.Request(f"{base_url}/models", headers=headers, method="GET")
+        try:
+            with urllib.request.urlopen(req, timeout=3) as resp:
+                data = json.loads(resp.read().decode("utf-8", errors="replace"))
+                items = data.get("data") or data.get("models") or []
+                if isinstance(items, list):
+                    for m in items:
+                        if isinstance(m, dict):
+                            m_id = str(m.get("id") or m.get("name") or "")
+                            for ck in ("context_window", "context_length", "max_model_len", "max_context_length", "n_ctx"):
+                                if ck in m and m[ck]:
+                                    try:
+                                        ctx_val = int(m[ck])
+                                        self._context_window_cache[f"{profile.profile_id}:{m_id}"] = ctx_val
+                                        self._context_window_cache[f"{profile.profile_id}:all"] = ctx_val
+                                        if not model or m_id == model or model in m_id or m_id in model or len(items) == 1:
+                                            return ctx_val
+                                    except (ValueError, TypeError):
+                                        pass
+                            meta = m.get("meta") or {}
+                            if isinstance(meta, dict):
+                                for ck in ("n_ctx", "context_length", "max_context_length"):
+                                    if ck in meta and meta[ck]:
+                                        try:
+                                            ctx_val = int(meta[ck])
+                                            self._context_window_cache[f"{profile.profile_id}:{m_id}"] = ctx_val
+                                            self._context_window_cache[f"{profile.profile_id}:all"] = ctx_val
+                                            if not model or m_id == model or model in m_id or m_id in model or len(items) == 1:
+                                                return ctx_val
+                                        except (ValueError, TypeError):
+                                            pass
+        except Exception as exc:
+            logger.debug("Failed to query context window from server for %s: %s", profile.profile_id, exc)
+
+        return None
+
     def invoke(self, profile: RouterProfileConfig, request: Dict[str, Any]) -> Dict[str, Any]:
         base_url = self._resolve_base_url(profile)
         api_key = self._resolve_api_key(profile)
@@ -54,9 +127,43 @@ class LocalLLMAdapter(BaseProviderAdapter):
         if not model or model == "default":
             model = profile.preferred_models[0] if profile.preferred_models else "default"
 
+        messages = list(request.get("messages", []))
+
+        # Context Truncation Guard: safely bound prompt if context_window is known to prevent VRAM overflow
+        context_window = self.get_context_window(profile, model, query_remote=False)
+        if context_window is not None and context_window > 0 and len(messages) > 1:
+            max_tok = int(request.get("max_tokens", 0) or 0)
+            token_budget = context_window - max_tok - 64
+            if token_budget > 100:
+                def _est_tok(msgs: list) -> int:
+                    total_chars = sum(len(str(m.get("content", ""))) for m in msgs if isinstance(m, dict))
+                    return int(total_chars / 3.5) + len(msgs) * 4
+
+                if _est_tok(messages) > token_budget:
+                    logger.warning(
+                        "Context truncation guard active for %s: prompt exceeds context window (%d). Truncating middle messages.",
+                        profile.profile_id,
+                        context_window,
+                    )
+                    system_msg = [messages[0]] if messages and messages[0].get("role") == "system" else []
+                    last_msg = messages[-1]
+                    middle = messages[1:-1] if system_msg else messages[:-1]
+
+                    while middle and _est_tok(system_msg + middle + [last_msg]) > token_budget:
+                        middle.pop(0)
+
+                    if _est_tok(system_msg + middle + [last_msg]) > token_budget:
+                        avail_chars = max(100, int(token_budget * 3.0))
+                        last_copy = dict(last_msg)
+                        last_copy["content"] = str(last_copy.get("content", ""))[-avail_chars:]
+                        messages = system_msg + middle + [last_copy]
+                    else:
+                        messages = system_msg + middle + [last_msg]
+
+
         payload: Dict[str, Any] = {
             "model": model,
-            "messages": request.get("messages", []),
+            "messages": messages,
             "temperature": request.get("temperature", 0.7),
         }
         if "tools" in request and request["tools"]:
@@ -105,6 +212,7 @@ class LocalLLMAdapter(BaseProviderAdapter):
         return data
 
 
+
     @staticmethod
     def _reject_empty_answer(data: Dict[str, Any]) -> None:
         """Пустой ответ — это отказ, а не успех.

src/antigravity_provider/router/auto_assigner.py

@@ -15,6 +15,7 @@ from pathlib import Path
 from typing import Any, Dict, List, Optional, Tuple
 
 from antigravity_provider.router.profile_manager import ProfileAuthManager, mask_email, mask_id
+from antigravity_provider.router.role_registry import RoleRegistry
 from antigravity_provider.router.router_config import (
     RolePolicy,
     RouterConfig,
@@ -25,19 +26,7 @@ from antigravity_provider.router.router_config import (
 
 logger = logging.getLogger("hermes.router.auto_assigner")
 
-HUMAN_ROLE_LABELS = {
-    "orchestrator_primary": "Главный оркестратор",
-    "orchestrator_fallback": "Резервный оркестратор",
-    "coder_1": "Кодер 1",
-    "coder_2": "Кодер 2",
-    "reviewer": "Ревьюер",
-    "researcher": "Исследователь",
-    "fast_agent": "Быстрый агент",
-    "universal_subagent": "Универсальный субагент",
-    "spare_1": "Резерв 1",
-    "spare_2": "Резерв 2",
-    "cold_spare": "Холодный резерв",
-}
+HUMAN_ROLE_LABELS = RoleRegistry.get_human_role_labels()
 
 DEFAULT_SLOT_ROLES = {
     "codex-orch": ("Главный оркестратор", "orchestrator", "primary"),
@@ -67,35 +56,7 @@ DEFAULT_SLOT_ROLES = {
 }
 
 
-CANONICAL_ROLE_MAP = {
-    "orchestrator": "orchestrator",
-    "orchestrator_primary": "orchestrator",
-    "orchestrator_fallback": "orchestrator",
-    "главный оркестратор": "orchestrator",
-    "резервный оркестратор": "orchestrator",
-    "coder": "coder-primary",
-    "coder_1": "coder-primary",
-    "coder-1": "coder-primary",
-    "coder-primary": "coder-primary",
-    "кодер 1": "coder-primary",
-    "coder_2": "coder-secondary",
-    "coder-2": "coder-secondary",
-    "coder-secondary": "coder-secondary",
-    "кодер 2": "coder-secondary",
-    "reviewer": "reviewer",
-    "ревьюер": "reviewer",
-    "research": "research",
-    "researcher": "research",
-    "исследователь": "research",
-    "fast": "fast",
-    "fast_agent": "fast",
-    "быстрый агент": "fast",
-    "general": "fast",
-    "universal_subagent": "fast",
-    "универсальный субагент": "fast",
-    "tester": "fast",
-    "тестировщик": "fast",
-}
+CANONICAL_ROLE_MAP = RoleRegistry.get_canonical_role_map()
 
 
 class AutoAssigner:
@@ -172,7 +133,7 @@ class AutoAssigner:
         candidates = list(provider_slots.get(provider_norm, []))
 
         # Priority based on requested role
-        if requested_role == "orchestrator":
+        if requested_role == "manager":
             if provider_norm in ("openai-codex", "codex") and "codex-orch" in candidates:
                 candidates.remove("codex-orch")
                 candidates.insert(0, "codex-orch")
@@ -368,7 +329,7 @@ class AutoAssigner:
                 "changes": [],
             }
 
-        canonical_roles = ["orchestrator", "coder-primary", "coder-secondary", "reviewer", "research", "fast"]
+        canonical_roles = [r for r in RoleRegistry.get_role_ids() if RoleRegistry.is_role_implemented(r)]
         changes: List[Dict[str, Any]] = []
 
         # Ensure canonical roles exist in config
@@ -418,12 +379,12 @@ class AutoAssigner:
         else:
             # Case 3: Multiple providers connected -> distribute by role provider preferences
             role_provider_preferences = {
-                "orchestrator": ["codex", "antigravity", "opencode", "claude", "grok", "local"],
-                "coder-primary": ["codex", "antigravity", "opencode", "claude", "grok", "local"],
-                "coder-secondary": ["codex", "antigravity", "opencode", "claude", "grok", "local"],
-                "reviewer": ["codex", "opencode", "antigravity", "claude", "grok", "local"],
-                "research": ["opencode", "antigravity", "grok", "claude", "codex", "local"],
-                "fast": ["opencode", "antigravity", "local", "grok", "codex", "claude"],
+                "manager": ["codex", "antigravity", "opencode", "claude", "grok", "local"],
+                "developer-1": ["codex", "antigravity", "opencode", "claude", "grok", "local"],
+                "developer-2": ["codex", "antigravity", "opencode", "claude", "grok", "local"],
+                "code-reviewer": ["codex", "opencode", "antigravity", "claude", "grok", "local"],
+                "researcher": ["opencode", "antigravity", "grok", "claude", "codex", "local"],
+                "tester": ["opencode", "antigravity", "local", "grok", "codex", "claude"],
             }
 
             by_prov: Dict[str, List[str]] = {}
@@ -503,7 +464,7 @@ class AutoAssigner:
         main_codex = ProfileAuthManager.get_main_profile("openai-codex")
 
         team = {
-            "orchestrator": [],
+            "manager": [],
             "subagents": [],
             "spares": [],
             "summary": {
@@ -625,3 +586,19 @@ class AutoAssigner:
             level="info",
         )
         return True, f"Цепочка роли '{canonical_role}' успешно сохранена: {', '.join(desired_chain)}"
+
+
+def ensure_profile_in_routing(profile_id: str) -> tuple[bool, str]:
+    """Keep existing chain rank or route a newly introduced profile slot.
+
+    Lives outside the GUI wizard so hermetic tests can import it without customtkinter.
+    """
+    config = load_router_config()
+    assigned_role = next(
+        (role_id for role_id, policy in config.roles.items() if profile_id in policy.preferred_chain),
+        "",
+    )
+    if assigned_role:
+        return True, f"Профиль уже входит в цепочку '{assigned_role}'"
+    _display_name, role_code, tier = AutoAssigner.get_display_name_and_role(profile_id)
+    return AutoAssigner.assign_profile_to_role(profile_id, role_code, is_primary=tier == "primary")

src/antigravity_provider/router/model_registry.py

@@ -417,6 +417,17 @@ class ModelRegistry:
             normalized = role.strip().lower()
             if normalized in self._role_reqs:
                 return self._role_reqs[normalized]
+            alias_map = {
+                "code-reviewer": "reviewer",
+                "manager": "orchestrator",
+                "developer-1": "coder-primary",
+                "developer-2": "coder-secondary",
+                "tester": "fast",
+                "researcher": "research",
+            }
+            mapped = alias_map.get(normalized)
+            if mapped and mapped in self._role_reqs:
+                return self._role_reqs[mapped]
             # Default fallback for custom roles
             return RoleRequirements(
                 role_id=normalized,

src/antigravity_provider/router/preflight_service.py

@@ -0,0 +1,394 @@
+"""Hermes Hub — Preflight Dependency & Readiness Check Service (Dependency Agent).
+
+Performs comprehensive zero-quota preflight validation of local environment, CLI tools,
+Python dependencies, local inference endpoints, role chain credentials, and disk permissions.
+"""
+from __future__ import annotations
+
+import importlib.util
+import logging
+import os
+import shutil
+import urllib.error
+import urllib.request
+from dataclasses import asdict, dataclass, field
+from typing import Any, Dict, List, Optional
+
+from antigravity_provider import paths
+from antigravity_provider.router.profile_manager import ProfileAuthManager
+from antigravity_provider.router.router_config import load_router_config
+
+logger = logging.getLogger("hermes.router.preflight")
+
+
+@dataclass
+class PreflightItem:
+    check_id: str
+    name: str
+    status: str  # "PASS" | "FAIL" | "WARN"
+    message: str
+    remediation: Optional[str] = None
+
+    def to_dict(self) -> Dict[str, Any]:
+        return asdict(self)
+
+
+@dataclass
+class PreflightReport:
+    success: bool
+    passed_count: int
+    failed_count: int
+    warn_count: int
+    checks: List[PreflightItem] = field(default_factory=list)
+
+    def to_dict(self) -> Dict[str, Any]:
+        return {
+            "success": self.success,
+            "passed_count": self.passed_count,
+            "failed_count": self.failed_count,
+            "warn_count": self.warn_count,
+            "checks": [c.to_dict() for c in self.checks],
+        }
+
+
+class PreflightCheckService:
+    """Zero-quota dependency and readiness inspection service."""
+
+    _instance: Optional[PreflightCheckService] = None
+
+    @classmethod
+    def get(cls) -> PreflightCheckService:
+        if cls._instance is None:
+            cls._instance = cls()
+        return cls._instance
+
+    def check_cli_dependencies(self) -> List[PreflightItem]:
+        """Check for external CLI executables and critical Python packages."""
+        items: List[PreflightItem] = []
+
+        # 1. Antigravity CLI (agy)
+        agy_path = shutil.which("agy") or shutil.which("agy.exe")
+        if agy_path:
+            items.append(
+                PreflightItem(
+                    check_id="cli_agy",
+                    name="CLI Antigravity (agy)",
+                    status="PASS",
+                    message=f"Исполняемый файл agy найден: {agy_path}",
+                )
+            )
+        else:
+            items.append(
+                PreflightItem(
+                    check_id="cli_agy",
+                    name="CLI Antigravity (agy)",
+                    status="WARN",
+                    message="Утилита 'agy' не найдена в системном PATH.",
+                    remediation="Установите agy CLI или добавьте каталог установки в системную переменную PATH.",
+                )
+            )
+
+        # 2. Python package: fastapi
+        fastapi_spec = importlib.util.find_spec("fastapi")
+        if fastapi_spec is not None:
+            items.append(
+                PreflightItem(
+                    check_id="pkg_fastapi",
+                    name="Библиотека FastAPI",
+                    status="PASS",
+                    message="Пакет fastapi успешно импортируется в окружении.",
+                )
+            )
+        else:
+            items.append(
+                PreflightItem(
+                    check_id="pkg_fastapi",
+                    name="Библиотека FastAPI",
+                    status="FAIL",
+                    message="Пакет 'fastapi' не установлен в текущем Python окружении.",
+                    remediation="Выполните 'pip install fastapi' для работы веб-интерфейса и REST API.",
+                )
+            )
+
+        # 3. Python package: uvicorn
+        uvicorn_spec = importlib.util.find_spec("uvicorn")
+        if uvicorn_spec is not None:
+            items.append(
+                PreflightItem(
+                    check_id="pkg_uvicorn",
+                    name="Библиотека Uvicorn",
+                    status="PASS",
+                    message="Пакет uvicorn успешно импортируется в окружении.",
+                )
+            )
+        else:
+            items.append(
+                PreflightItem(
+                    check_id="pkg_uvicorn",
+                    name="Библиотека Uvicorn",
+                    status="FAIL",
+                    message="Пакет 'uvicorn' не установлен в текущем Python окружении.",
+                    remediation="Выполните 'pip install uvicorn' для запуска веб-сервера.",
+                )
+            )
+
+        return items
+
+    def check_local_servers(self) -> List[PreflightItem]:
+        """Poll {base_url}/models with 2.0s timeout for active local provider profiles."""
+        items: List[PreflightItem] = []
+        config = load_router_config()
+        local_profiles = [p for p in config.profiles.values() if p.provider == "local" and p.enabled]
+
+        if not local_profiles:
+            items.append(
+                PreflightItem(
+                    check_id="local_servers_none",
+                    name="Локальные серверы LLM",
+                    status="PASS",
+                    message="Активные локальные профили (llama.cpp/vLLM) не настроены.",
+                )
+            )
+            return items
+
+        from antigravity_provider.router.adapters.local_adapter import LocalLLMAdapter
+
+        adapter = LocalLLMAdapter()
+
+        for pcfg in local_profiles:
+            base_url = adapter._resolve_base_url(pcfg)
+            api_key = adapter._resolve_api_key(pcfg)
+            headers = {"Accept": "application/json", "User-Agent": "hermes-preflight/1.0"}
+            if api_key:
+                headers["Authorization"] = f"Bearer {api_key}"
+
+            models_url = f"{base_url}/models"
+            try:
+                req = urllib.request.Request(models_url, headers=headers, method="GET")
+                with urllib.request.urlopen(req, timeout=2.0) as resp:
+                    if resp.status in (200, 204):
+                        items.append(
+                            PreflightItem(
+                                check_id=f"local_srv_{pcfg.profile_id}",
+                                name=f"Локальный сервер {pcfg.profile_id} ({base_url})",
+                                status="PASS",
+                                message=f"Локальный сервер доступен (HTTP {resp.status}).",
+                            )
+                        )
+                    else:
+                        items.append(
+                            PreflightItem(
+                                check_id=f"local_srv_{pcfg.profile_id}",
+                                name=f"Локальный сервер {pcfg.profile_id} ({base_url})",
+                                status="FAIL",
+                                message=f"Сервер вернул неожиданный статус HTTP {resp.status}",
+                                remediation=f"Проверьте настройки и логи сервера {base_url}.",
+                            )
+                        )
+            except urllib.error.HTTPError as http_err:
+                items.append(
+                    PreflightItem(
+                        check_id=f"local_srv_{pcfg.profile_id}",
+                        name=f"Локальный сервер {pcfg.profile_id} ({base_url})",
+                        status="FAIL",
+                        message=f"HTTP ошибка при обращении к {models_url}: {http_err.code} {http_err.reason}",
+                        remediation=f"Убедитесь, что сервер на {base_url} поддерживает OpenAI-совместимый эндпоинт /v1/models.",
+                    )
+                )
+            except Exception as exc:
+                items.append(
+                    PreflightItem(
+                        check_id=f"local_srv_{pcfg.profile_id}",
+                        name=f"Локальный сервер {pcfg.profile_id} ({base_url})",
+                        status="FAIL",
+                        message=f"Не удалось подключиться к {base_url}: {exc}",
+                        remediation=f"Запустите локальный сервер llama.cpp / vLLM / Ollama по адресу {base_url}.",
+                    )
+                )
+
+        return items
+
+    def check_auth_credentials(self) -> List[PreflightItem]:
+        """Verify credential presence for all profiles referenced in active role chains.
+
+        ZERO QUOTA BURN: Only inspects local auth files and keyring status. Never calls paid APIs.
+        """
+        items: List[PreflightItem] = []
+        config = load_router_config()
+
+        # Collect all profile IDs in active role chains
+        referenced_pids: set[str] = set()
+        for role_policy in config.roles.values():
+            for pid in role_policy.preferred_chain:
+                referenced_pids.add(pid)
+
+        if not referenced_pids:
+            items.append(
+                PreflightItem(
+                    check_id="auth_chains_empty",
+                    name="Учетные данные цепочек ролей",
+                    status="WARN",
+                    message="В активных ролях не настроены цепочки профилей.",
+                    remediation="Настройте цепочки профилей в разделе Маршрутизация.",
+                )
+            )
+            return items
+
+        for pid in sorted(referenced_pids):
+            pcfg = config.get_profile(pid)
+            if not pcfg:
+                items.append(
+                    PreflightItem(
+                        check_id=f"auth_{pid}",
+                        name=f"Профиль {pid}",
+                        status="FAIL",
+                        message=f"Профиль '{pid}' указан в цепочке роли, но отсутствует в конфигурации.",
+                        remediation=f"Удалите '{pid}' из цепочки роли или настройте профиль в router_profiles.yaml.",
+                    )
+                )
+                continue
+
+            status = ProfileAuthManager.get_profile_status(pcfg.provider, pid)
+            is_authenticated = status.get("authenticated", False)
+            is_expired = status.get("is_expired", False) or status.get("expired", False) or status.get("status") == "EXPIRED"
+
+            if is_authenticated and not is_expired:
+                items.append(
+                    PreflightItem(
+                        check_id=f"auth_{pid}",
+                        name=f"Авторизация {pid} ({pcfg.provider})",
+                        status="PASS",
+                        message="Учетные данные действительны и сохранены локально.",
+                    )
+                )
+            elif is_expired:
+                items.append(
+                    PreflightItem(
+                        check_id=f"auth_{pid}",
+                        name=f"Авторизация {pid} ({pcfg.provider})",
+                        status="FAIL",
+                        message=f"Срок действия авторизации для профиля '{pid}' истек.",
+                        remediation=f"Выполните повторный вход для профиля {pid} в разделе Аккаунты.",
+                    )
+                )
+            else:
+                items.append(
+                    PreflightItem(
+                        check_id=f"auth_{pid}",
+                        name=f"Авторизация {pid} ({pcfg.provider})",
+                        status="FAIL",
+                        message=f"Учетные данные для профиля '{pid}' ({pcfg.provider}) не найдены.",
+                        remediation=f"Подключите профиль {pid} через кнопку 'Добавить аккаунт' или 'hermes router login'.",
+                    )
+                )
+
+        return items
+
+    def check_system_environment(self) -> List[PreflightItem]:
+        """Verify HERMES_HOME presence and read/write permissions for config and logs."""
+        items: List[PreflightItem] = []
+
+        # 1. HERMES_HOME directory
+        try:
+            home_dir = paths.get_hermes_home()
+            if home_dir.is_dir():
+                items.append(
+                    PreflightItem(
+                        check_id="env_hermes_home",
+                        name="Каталог HERMES_HOME",
+                        status="PASS",
+                        message=f"Каталог существует: {home_dir}",
+                    )
+                )
+            else:
+                items.append(
+                    PreflightItem(
+                        check_id="env_hermes_home",
+                        name="Каталог HERMES_HOME",
+                        status="FAIL",
+                        message=f"Каталог {home_dir} не существует или не является директорией.",
+                        remediation="Проверьте права доступа и создайте каталог HERMES_HOME.",
+                    )
+                )
+        except Exception as exc:
+            items.append(
+                PreflightItem(
+                    check_id="env_hermes_home",
+                    name="Каталог HERMES_HOME",
+                    status="FAIL",
+                    message=f"Ошибка доступа к HERMES_HOME: {exc}",
+                    remediation="Убедитесь, что переменная HERMES_HOME указывает на корректный доступный путь.",
+                )
+            )
+
+        # 2. Config Directory Write Test
+        try:
+            config_dir = paths.get_config_dir()
+            test_file = config_dir / f".preflight_probe_{os.getpid()}.tmp"
+            test_file.write_text("probe", encoding="utf-8")
+            test_file.unlink()
+            items.append(
+                PreflightItem(
+                    check_id="env_config_writable",
+                    name="Права на запись в каталог конфигурации",
+                    status="PASS",
+                    message=f"Права на запись в {config_dir} подтверждены.",
+                )
+            )
+        except Exception as exc:
+            items.append(
+                PreflightItem(
+                    check_id="env_config_writable",
+                    name="Права на запись в каталог конфигурации",
+                    status="FAIL",
+                    message=f"Нет прав на запись в {paths.get_config_dir()}: {exc}",
+                    remediation="Предоставьте текущему пользователю права на запись в каталог конфигурации.",
+                )
+            )
+
+        # 3. Logs Directory Write Test
+        try:
+            logs_dir = paths.get_logs_dir()
+            test_file = logs_dir / f".preflight_probe_{os.getpid()}.tmp"
+            test_file.write_text("probe", encoding="utf-8")
+            test_file.unlink()
+            items.append(
+                PreflightItem(
+                    check_id="env_logs_writable",
+                    name="Права на запись в каталог логов",
+                    status="PASS",
+                    message=f"Права на запись в {logs_dir} подтверждены.",
+                )
+            )
+        except Exception as exc:
+            items.append(
+                PreflightItem(
+                    check_id="env_logs_writable",
+                    name="Права на запись в каталог логов",
+                    status="FAIL",
+                    message=f"Нет прав на запись в {paths.get_logs_dir()}: {exc}",
+                    remediation="Предоставьте текущему пользователю права на запись в каталог логов.",
+                )
+            )
+
+        return items
+
+    def run_all_checks(self) -> PreflightReport:
+        """Run all readiness checks and return aggregated PreflightReport."""
+        all_items: List[PreflightItem] = []
+        all_items.extend(self.check_cli_dependencies())
+        all_items.extend(self.check_system_environment())
+        all_items.extend(self.check_auth_credentials())
+        all_items.extend(self.check_local_servers())
+
+        passed = sum(1 for item in all_items if item.status == "PASS")
+        failed = sum(1 for item in all_items if item.status == "FAIL")
+        warn = sum(1 for item in all_items if item.status == "WARN")
+
+        return PreflightReport(
+            success=(failed == 0),
+            passed_count=passed,
+            failed_count=failed,
+            warn_count=warn,
+            checks=all_items,
+        )

src/antigravity_provider/router/role_registry.py

@@ -0,0 +1,357 @@
+"""Hermes Multi-Provider Account Router — Unified Role Registry."""
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from typing import TYPE_CHECKING, Any, Dict, List, Optional, Tuple
+
+if TYPE_CHECKING:
+    from antigravity_provider.router.router_config import RolePolicy
+
+
+
+STATUS_DECLARED_UNIMPLEMENTED = "DECLARED_UNIMPLEMENTED"
+STATUS_DECLARED_UNIMPLEMENTED_LABEL_RU = "Роль объявлена, исполнение не реализовано"
+
+@dataclass(frozen=True)
+class RoleDefinition:
+    role_id: str
+    display_name_ru: str
+    short_name_ru: str
+    description_ru: str
+    is_implemented: bool = True
+    unimplemented_reason_ru: Optional[str] = None
+    capabilities: List[str] = field(default_factory=list)
+    fallback_capabilities: List[str] = field(default_factory=list)
+    default_preferred_chain: List[str] = field(default_factory=list)
+    default_model: Optional[str] = None
+    max_failover_attempts: int = 3
+    session_affinity_enabled: bool = True
+    is_orchestrator: bool = False
+    tier: str = "core"
+
+CANONICAL_ROLES: Dict[str, RoleDefinition] = {
+    "manager": RoleDefinition(
+        role_id="manager",
+        display_name_ru="Менеджер проекта (Оркестратор)",
+        short_name_ru="Менеджер",
+        description_ru="Координация работы субагентов, планирование, контроль исполнения и распределение задач.",
+        is_implemented=True,
+        is_orchestrator=True,
+        capabilities=["orchestrator", "reasoning", "management", "planning"],
+        fallback_capabilities=["orchestrator", "reasoning"],
+        default_preferred_chain=["codex-orch", "ag-orch-fallback", "claude-orch", "grok-orch", "opengo-3"],
+        default_model="gemini-3.7-flash",
+        max_failover_attempts=3,
+        tier="core",
+    ),
+    "developer-1": RoleDefinition(
+        role_id="developer-1",
+        display_name_ru="Ведущий разработчик (Кодер 1)",
+        short_name_ru="Кодер 1",
+        description_ru="Пишет код, реализует функционал, исправляет ошибки.",
+        is_implemented=True,
+        capabilities=["coding", "developer-1", "reasoning", "tools", "structured_output"],
+        fallback_capabilities=["coding"],
+        default_preferred_chain=["codex-worker-1", "claude-worker-1", "ag-w1", "opengo-1"],
+        max_failover_attempts=3,
+        tier="core",
+    ),
+    "developer-2": RoleDefinition(
+        role_id="developer-2",
+        display_name_ru="Вспомогательный разработчик (Кодер 2)",
+        short_name_ru="Кодер 2",
+        description_ru="Проверяет код Разработчика 1 и выдаёт ему задание на исправление.",
+        is_implemented=True,
+        capabilities=["coding", "developer-2", "reviewer", "tools"],
+        fallback_capabilities=["coding", "reviewer"],
+        default_preferred_chain=["ag-w1", "grok-worker-1", "codex-worker-2", "opengo-3"],
+        max_failover_attempts=3,
+        tier="core",
+    ),
+    "code-reviewer": RoleDefinition(
+        role_id="code-reviewer",
+        display_name_ru="Ревьюер кода",
+        short_name_ru="Код-ревьювер",
+        description_ru="Анализирует код на ошибки, проблемы безопасности и соответствие стандартам. Работает после Разработчика 2.",
+        is_implemented=True,
+        capabilities=["code-reviewer", "reviewer", "coding", "security_analysis"],
+        fallback_capabilities=["reviewer", "coding"],
+        default_preferred_chain=["claude-worker-2", "codex-worker-2", "ag-w2", "opengo-2"],
+        max_failover_attempts=3,
+        tier="core",
+    ),
+    "researcher": RoleDefinition(
+        role_id="researcher",
+        display_name_ru="Исследователь",
+        short_name_ru="Исследователь",
+        description_ru="Изучает данные, кодовую базу, документацию и внешние источники, чтобы собрать информацию для решения задачи.",
+        is_implemented=True,
+        capabilities=["researcher", "research", "search", "long_context"],
+        fallback_capabilities=["research", "search"],
+        default_preferred_chain=["ag-w3", "grok-worker-2", "opengo-2", "opengo-1"],
+        max_failover_attempts=3,
+        tier="core",
+    ),
+    "tester": RoleDefinition(
+        role_id="tester",
+        display_name_ru="Тестировщик (QA)",
+        short_name_ru="Тестировщик",
+        description_ru="Создаёт тесты, проверяет корректность работы кода, находит дефекты.",
+        is_implemented=True,
+        capabilities=["tester", "testing", "fast", "automation"],
+        fallback_capabilities=["fast", "testing"],
+        default_preferred_chain=["ag-w4", "opengo-1", "ag-spare-1"],
+        max_failover_attempts=3,
+        tier="qa_doc",
+    ),
+    "tech-writer": RoleDefinition(
+        role_id="tech-writer",
+        display_name_ru="Технический писатель",
+        short_name_ru="Техписатель",
+        description_ru="Создаёт документацию, инструкции, README.",
+        is_implemented=True,
+        capabilities=["tech-writer", "documentation", "reasoning", "structured_output"],
+        fallback_capabilities=["documentation", "reasoning"],
+        default_preferred_chain=["claude-worker-2", "ag-w3", "ag-w2"],
+        max_failover_attempts=3,
+        tier="qa_doc",
+    ),
+    "analyst": RoleDefinition(
+        role_id="analyst",
+        display_name_ru="Системный аналитик",
+        short_name_ru="Аналитик",
+        description_ru="Проводит глубокий анализ данных, выявляет тренды, строит прогнозы.",
+        is_implemented=True,
+        capabilities=["analyst", "reasoning", "research", "planning"],
+        fallback_capabilities=["reasoning", "research"],
+        default_preferred_chain=["grok-worker-2", "opengo-2", "ag-w3"],
+        max_failover_attempts=3,
+        tier="qa_doc",
+    ),
+    "guardian": RoleDefinition(
+        role_id="guardian",
+        display_name_ru="Надзиратель (агент безопасности)",
+        short_name_ru="Надзиратель",
+        description_ru="Проверяет входящие инструкции на промпт-инъекции, анализирует планы и вызовы инструментов, блокирует обход системных правил, не допускает утечки секретов, следит за границами песочницы.",
+        is_implemented=False,
+        unimplemented_reason_ru=STATUS_DECLARED_UNIMPLEMENTED_LABEL_RU,
+        capabilities=["guardian", "security", "sandboxing"],
+        fallback_capabilities=["guardian", "security"],
+        default_preferred_chain=[],
+        max_failover_attempts=0,
+        tier="governance",
+    ),
+    "cost-controller": RoleDefinition(
+        role_id="cost-controller",
+        display_name_ru="Агент контроля затрат",
+        short_name_ru="Контроль затрат",
+        description_ru="Оценивает планируемый расход токенов, сравнивает с остатком бюджета, предлагает упрощения, сверяет факт с прогнозом, останавливает цепочку при исчерпании лимита.",
+        is_implemented=False,
+        unimplemented_reason_ru=STATUS_DECLARED_UNIMPLEMENTED_LABEL_RU,
+        capabilities=["cost-controller", "budget", "analytics"],
+        fallback_capabilities=["cost-controller", "budget"],
+        default_preferred_chain=[],
+        max_failover_attempts=0,
+        tier="governance",
+    ),
+    "integration-expert": RoleDefinition(
+        role_id="integration-expert",
+        display_name_ru="Специалист по интеграции",
+        short_name_ru="Интегратор",
+        description_ru="Работает с API и внешними сервисами, отправляет вебхуки.",
+        is_implemented=True,
+        capabilities=["integration-expert", "integration", "coding", "networking"],
+        fallback_capabilities=["integration", "coding"],
+        default_preferred_chain=["opengo-3", "codex-worker-1", "ag-w4"],
+        max_failover_attempts=3,
+        tier="expert",
+    ),
+    "security-expert": RoleDefinition(
+        role_id="security-expert",
+        display_name_ru="Юрист / специалист по безопасности",
+        short_name_ru="Безопасник",
+        description_ru="Проверяет код и данные на уязвимости.",
+        is_implemented=True,
+        capabilities=["security-expert", "security", "code-reviewer", "audit"],
+        fallback_capabilities=["security", "code-reviewer"],
+        default_preferred_chain=["claude-worker-1", "codex-worker-2", "ag-w2"],
+        max_failover_attempts=3,
+        tier="expert",
+    ),
+    "dependency-agent": RoleDefinition(
+        role_id="dependency-agent",
+        display_name_ru="Проверяющий готовность",
+        short_name_ru="Готовность",
+        description_ru="До начала задачи убеждается, что на месте всё необходимое — исполняемые файлы и CLI, библиотеки, учётные данные, права доступа, доступность локальных серверов. Сообщает о нехватке до запуска.",
+        is_implemented=True,
+        capabilities=["dependency-agent", "preflight", "environment", "system_checks", "fast"],
+        fallback_capabilities=["dependency-agent", "preflight"],
+        default_preferred_chain=["opengo-1", "ag-w1", "codex-worker-1"],
+        max_failover_attempts=3,
+        tier="qa_doc",
+    ),
+}
+
+_CANONICAL_ROLE_ALIASES: Dict[str, str] = {
+    "orchestrator": "manager",
+    "главный оркестратор": "manager",
+    "оркестратор": "manager",
+    "менеджер": "manager",
+    "coder": "developer-1",
+    "coder-primary": "developer-1",
+    "developer": "developer-1",
+    "кодер": "developer-1",
+    "кодер 1": "developer-1",
+    "разработчик": "developer-1",
+    "разработчик 1": "developer-1",
+    "coder-secondary": "developer-2",
+    "кодер 2": "developer-2",
+    "разработчик 2": "developer-2",
+    "reviewer": "code-reviewer",
+    "ревьюер": "code-reviewer",
+    "код-ревьювер": "code-reviewer",
+    "код-ревьюер": "code-reviewer",
+    "research": "researcher",
+    "исследователь": "researcher",
+    "fast": "tester",
+    "general": "tester",
+    "тестировщик": "tester",
+    "быстрый агент": "tester",
+    "tech_writer": "tech-writer",
+    "технический писатель": "tech-writer",
+    "аналитик": "analyst",
+    "надзиратель": "guardian",
+    "контроль затрат": "cost-controller",
+    "агент контроля затрат": "cost-controller",
+    "интеграция": "integration-expert",
+    "специалист по интеграции": "integration-expert",
+    "безопасность": "security-expert",
+    "специалист по безопасности": "security-expert",
+    "dependency-agent": "dependency-agent",
+    "dependency_agent": "dependency-agent",
+    "preflight": "dependency-agent",
+    "проверяющий готовность": "dependency-agent",
+    "агент зависимостей": "dependency-agent",
+    "готовность": "dependency-agent",
+    "dependency": "dependency-agent",
+}
+
+class RoleRegistry:
+    @classmethod
+    def get_all_roles(cls) -> Dict[str, RoleDefinition]:
+        return dict(CANONICAL_ROLES)
+
+    @classmethod
+    def get_role(cls, role_id: str) -> Optional[RoleDefinition]:
+        canonical_id = cls.resolve_canonical_role(role_id)
+        return CANONICAL_ROLES.get(canonical_id)
+
+    @classmethod
+    def get_role_ids(cls) -> List[str]:
+        return list(CANONICAL_ROLES.keys())
+
+    @classmethod
+    def get_executable_role_ids(cls) -> List[str]:
+        return [r_id for r_id, r_def in CANONICAL_ROLES.items() if r_def.is_implemented]
+
+    @classmethod
+    def get_unimplemented_role_ids(cls) -> List[str]:
+        return [r_id for r_id, r_def in CANONICAL_ROLES.items() if not r_def.is_implemented]
+
+    @classmethod
+    def is_role_implemented(cls, role_id: str) -> bool:
+        r_def = cls.get_role(role_id)
+        return r_def.is_implemented if r_def else False
+
+    @classmethod
+    def get_role_name_ru(cls, role_id: str, default: Optional[str] = None) -> str:
+        r_def = cls.get_role(role_id)
+        if r_def:
+            return r_def.display_name_ru
+        return default or role_id
+
+    @classmethod
+    def get_role_short_name_ru(cls, role_id: str, default: Optional[str] = None) -> str:
+        r_def = cls.get_role(role_id)
+        if r_def:
+            return r_def.short_name_ru
+        return default or role_id
+
+    @classmethod
+    def get_role_description_ru(cls, role_id: str) -> str:
+        r_def = cls.get_role(role_id)
+        if r_def:
+            return r_def.description_ru
+        return ""
+
+    @classmethod
+    def resolve_canonical_role(cls, name_or_alias: str) -> str:
+        if not name_or_alias:
+            return "manager"
+        clean = name_or_alias.strip().lower()
+        return _CANONICAL_ROLE_ALIASES.get(clean, clean)
+
+    @classmethod
+    def resolve_role_name(cls, name_or_alias: str) -> str:
+        return cls.resolve_canonical_role(name_or_alias)
+
+    @classmethod
+    def get_canonical_role_map(cls) -> Dict[str, str]:
+        return dict(_CANONICAL_ROLE_ALIASES)
+
+    @classmethod
+    def get_human_role_labels(cls) -> Dict[str, str]:
+        labels = {}
+        for r_id, r_def in CANONICAL_ROLES.items():
+            labels[r_id] = r_def.short_name_ru
+        labels.update({
+            "spare_1": "Резерв 1",
+            "spare_2": "Резерв 2",
+            "cold_spare": "Холодный резерв",
+        })
+        return labels
+
+    @classmethod
+    def get_default_role_policies(cls) -> Dict[str, "RolePolicy"]:
+        from antigravity_provider.router.router_config import RolePolicy
+        policies: Dict[str, "RolePolicy"] = {}
+        for r_id, r_def in CANONICAL_ROLES.items():
+            policies[r_id] = RolePolicy(
+                role_name=r_id,
+                preferred_chain=list(r_def.default_preferred_chain),
+                fallback_capabilities=list(r_def.fallback_capabilities),
+                max_failover_attempts=r_def.max_failover_attempts,
+                session_affinity_enabled=r_def.session_affinity_enabled,
+                default_model=r_def.default_model,
+            )
+        return policies
+
+    @classmethod
+    def migrate_legacy_roles(cls, current_roles: Dict[str, "RolePolicy"]) -> Tuple[Dict[str, "RolePolicy"], bool]:
+        from antigravity_provider.router.router_config import RolePolicy
+        migrated: Dict[str, "RolePolicy"] = {}
+        was_modified = False
+        default_policies = cls.get_default_role_policies()
+
+        for rname, rpol in current_roles.items():
+            canonical_id = cls.resolve_canonical_role(rname)
+            if canonical_id != rname:
+                was_modified = True
+                if canonical_id not in migrated:
+                    migrated[canonical_id] = RolePolicy(
+                        role_name=canonical_id,
+                        preferred_chain=list(rpol.preferred_chain),
+                        fallback_capabilities=list(rpol.fallback_capabilities) or list(default_policies[canonical_id].fallback_capabilities),
+                        max_failover_attempts=rpol.max_failover_attempts,
+                        session_affinity_enabled=rpol.session_affinity_enabled,
+                        default_model=rpol.default_model or default_policies[canonical_id].default_model,
+                    )
+            else:
+                migrated[rname] = rpol
+
+        for canon_id, def_policy in default_policies.items():
+            if canon_id not in migrated:
+                migrated[canon_id] = def_policy
+                was_modified = True
+
+        return migrated, was_modified

src/antigravity_provider/router/router_config.py

@@ -3,6 +3,7 @@ from __future__ import annotations
 
 import os
 from dataclasses import dataclass, field
+from antigravity_provider.router.role_registry import RoleRegistry
 from pathlib import Path
 from typing import Any, Dict, List, Optional
 import yaml
@@ -258,7 +259,7 @@ def get_default_router_config() -> RouterConfig:
             profile_id="local-1",
             provider="local",
             account_id="local-acc-1",
-            capabilities=["reviewer", "coder-secondary", "reasoning", "coding"],
+            capabilities=["code-reviewer", "reviewer", "coder-secondary", "reasoning", "coding"],
             preferred_models=["Qwen3.8-27B-Q4_K_M.gguf", "default"],
             max_concurrency=1,
         ),
@@ -266,57 +267,13 @@ def get_default_router_config() -> RouterConfig:
             profile_id="local-2",
             provider="local",
             account_id="local-acc-2",
-            capabilities=["fast", "research", "coding"],
+            capabilities=["tester", "fast", "research", "coding"],
             preferred_models=["Qwen3-4B-Instruct-2507-Q4_K_M.gguf", "default"],
             max_concurrency=1,
         ),
     }
 
-    roles: dict[str, RolePolicy] = {
-        "orchestrator": RolePolicy(
-            role_name="orchestrator",
-            preferred_chain=["codex-orch", "ag-orch-fallback", "opengo-3"],
-            fallback_capabilities=["orchestrator", "reasoning"],
-            max_failover_attempts=3,
-            session_affinity_enabled=True,
-            default_model="gemini-3.7-flash",
-        ),
-        "coder-primary": RolePolicy(
-            role_name="coder-primary",
-            preferred_chain=["codex-worker-1", "ag-w1", "opengo-3"],
-            fallback_capabilities=["coding"],
-            max_failover_attempts=3,
-            session_affinity_enabled=True,
-        ),
-        "coder-secondary": RolePolicy(
-            role_name="coder-secondary",
-            preferred_chain=["codex-worker-2", "ag-w2", "opengo-2"],
-            fallback_capabilities=["coding", "reviewer"],
-            max_failover_attempts=3,
-            session_affinity_enabled=True,
-        ),
-        "reviewer": RolePolicy(
-            role_name="reviewer",
-            preferred_chain=["codex-worker-2", "opengo-2", "ag-w2"],
-            fallback_capabilities=["reviewer", "coding"],
-            max_failover_attempts=3,
-            session_affinity_enabled=True,
-        ),
-        "research": RolePolicy(
-            role_name="research",
-            preferred_chain=["opengo-1", "ag-w3", "ag-w4"],
-            fallback_capabilities=["research", "search"],
-            max_failover_attempts=3,
-            session_affinity_enabled=True,
-        ),
-        "fast": RolePolicy(
-            role_name="fast",
-            preferred_chain=["opengo-1", "ag-w4", "ag-spare-1"],
-            fallback_capabilities=["fast"],
-            max_failover_attempts=3,
-            session_affinity_enabled=True,
-        ),
-    }
+    roles = RoleRegistry.get_default_role_policies()
 
     return RouterConfig(
         enabled=True,
@@ -346,16 +303,20 @@ def load_router_config(config_path: Optional[Path] = None) -> RouterConfig:
         profiles_raw = data.get("profiles", {})
         profiles: dict[str, RouterProfileConfig] = {}
         for pid, pdata in profiles_raw.items():
+            provider = pdata.get("provider", "antigravity")
+            max_concurrency = int(pdata.get("max_concurrency", 1))
+            if provider == "local":
+                max_concurrency = 1
             profiles[pid] = RouterProfileConfig(
                 profile_id=pid,
-                provider=pdata.get("provider", "antigravity"),
+                provider=provider,
                 account_id=pdata.get("account_id", pid),
                 capabilities=list(pdata.get("capabilities", [])),
                 preferred_models=list(pdata.get("preferred_models", [])),
                 fallback_models=list(pdata.get("fallback_models", [])),
                 auth_config=dict(pdata.get("auth_config", {})),
                 enabled=bool(pdata.get("enabled", True)),
-                max_concurrency=int(pdata.get("max_concurrency", 1)),
+                max_concurrency=max_concurrency,
                 custom_base_url=pdata.get("custom_base_url"),
             )
 

src/antigravity_provider/router/settings_service.py

@@ -22,9 +22,11 @@ DEFAULT_SETTINGS: Dict[str, Any] = {
     "monitoring_interval_seconds": 30,
     "quota_threshold_percent": 10.0,
     "quota_threshold_action": "notify",
+    "email_masking_mode": "none",
 }
 
 
+
 _SETTINGS_CACHE: Dict[str, Any] | None = None
 _SETTINGS_CACHE_MTIME: float = -1.0
 _SETTINGS_CACHE_PATH: str = ""
@@ -98,6 +100,11 @@ def get_hub_settings() -> Dict[str, Any]:
         action = "notify"
     merged["quota_threshold_action"] = action
 
+    email_mode = str(merged.get("email_masking_mode", "none")).strip().lower()
+    if email_mode not in ("none", "partial", "full"):
+        email_mode = "none"
+    merged["email_masking_mode"] = email_mode
+
     _SETTINGS_CACHE = dict(merged)
     _SETTINGS_CACHE_MTIME = current_mtime
     _SETTINGS_CACHE_PATH = sfile_str

src/antigravity_provider/router/state_store.py

@@ -55,6 +55,7 @@ class HubSnapshot:
     quotas: Dict[str, Any]
     metrics: Dict[str, Any] = field(default_factory=dict)
     is_stale: bool = False
+    workflow: Dict[str, Any] = field(default_factory=dict)
 
     def get_profile(self, profile_id: str) -> Optional[ProfileViewModel]:
         return self.all_profiles.get(profile_id)
@@ -107,6 +108,24 @@ class HubStateStore:
             if self._current_snapshot is not None:
                 if (time.time() - self._current_snapshot.timestamp > 300.0) and not self._current_snapshot.is_stale:
                     self._current_snapshot = replace(self._current_snapshot, is_stale=True)
+                # Provider/account scans are intentionally cached, while LIVE
+                # workflow checkpoints are small local state and must never lag
+                # behind an action until the next expensive provider refresh.
+                try:
+                    from .workflow_service import WorkflowService
+
+                    live_workflow = WorkflowService.get().snapshot()
+                    role_views = {agent.role_id: agent for agent in self._current_snapshot.agents}
+                    for workflow_agent in live_workflow.get("agents", []):
+                        role_view = role_views.get(workflow_agent.get("role"))
+                        generic_name = str(workflow_agent.get("role") or "").replace("-", " ").title()
+                        if role_view and workflow_agent.get("name") == generic_name:
+                            workflow_agent["name"] = role_view.role_name_ru
+                        if role_view and not workflow_agent.get("description"):
+                            workflow_agent["description"] = role_view.role_description_ru
+                    self._current_snapshot = replace(self._current_snapshot, workflow=live_workflow)
+                except Exception:
+                    pass
                 return self._current_snapshot
         return self.refresh(force_scan=False)
 
@@ -196,6 +215,27 @@ class HubStateStore:
                 "active_calls_total": active_leases_total,
                 "active_calls_by_profile": active_leases_by_profile,
             }
+            try:
+                from .workflow_service import WorkflowService
+
+                workflow_data = WorkflowService.get().snapshot()
+                role_views = {agent.role_id: agent for agent in agents}
+                for workflow_agent in workflow_data.get("agents", []):
+                    role_view = role_views.get(workflow_agent.get("role"))
+                    generic_name = str(workflow_agent.get("role") or "").replace("-", " ").title()
+                    if role_view and workflow_agent.get("name") == generic_name:
+                        workflow_agent["name"] = role_view.role_name_ru
+                    if role_view and not workflow_agent.get("description"):
+                        workflow_agent["description"] = role_view.role_description_ru
+            except Exception as exc:
+                workflow_data = {
+                    "agents": [],
+                    "definition": {},
+                    "run": {"status": "unavailable"},
+                    "events": [],
+                    "is_loading": False,
+                    "unavailable_reason": f"Workflow state unavailable: {exc}",
+                }
             snapshot = HubSnapshot(
                 generation=gen,
                 seq=request_seq,
@@ -209,6 +249,7 @@ class HubStateStore:
                 quotas=quotas_map,
                 metrics=metrics,
                 is_stale=False,
+                workflow=workflow_data,
             )
             self._current_snapshot = snapshot
 
@@ -263,6 +304,13 @@ class HubStateStore:
                 "active_calls_by_profile": {},
             },
             is_stale=True,
+            workflow={
+                "agents": [],
+                "definition": {},
+                "run": {"status": "loading"},
+                "events": [],
+                "is_loading": True,
+            },
         )
 
     def _apply_profile_delta(self, profile: ProfileViewModel) -> HubSnapshot:

src/antigravity_provider/router/telemetry_service.py

@@ -8,6 +8,7 @@ Captures, persists, and computes honest empirical measurements for all router ca
 - Ring Buffer & File Rotation (bounded memory and disk footprint)
 """
 from __future__ import annotations
+from antigravity_provider.router.role_registry import RoleRegistry
 
 import collections
 import datetime
@@ -30,6 +31,15 @@ MAX_FILE_BYTES = 5 * 1024 * 1024  # 5 MB
 MAX_BACKUP_FILES = 3
 
 
+def format_token_count(measured: Optional[int], estimated: Optional[int]) -> Optional[str]:
+    """Format token count distinguishing measured exact counts from estimations (~)."""
+    if measured is not None:
+        return str(measured)
+    elif estimated is not None:
+        return f"~{estimated}"
+    return None
+
+
 @dataclass
 class TelemetryRecord:
     """Immutable record of an individual router invocation attempt."""
@@ -44,6 +54,13 @@ class TelemetryRecord:
     prompt_tokens: Optional[int] = None
     completion_tokens: Optional[int] = None
     total_tokens: Optional[int] = None
+    prompt_tokens_measured: Optional[int] = None
+    prompt_tokens_estimated: Optional[int] = None
+    completion_tokens_measured: Optional[int] = None
+    completion_tokens_estimated: Optional[int] = None
+    total_tokens_measured: Optional[int] = None
+    total_tokens_estimated: Optional[int] = None
+    is_estimated: bool = False
     cost_usd: Optional[float] = None
     failover_count: int = 0
     error_category: Optional[str] = None
@@ -68,6 +85,14 @@ class TelemetryAggregates:
     total_prompt_tokens: Optional[int] = None   # Sum of reported prompt tokens or None if no token data
     total_completion_tokens: Optional[int] = None
     total_tokens: Optional[int] = None
+    total_prompt_tokens_measured: Optional[int] = None
+    total_prompt_tokens_estimated: Optional[int] = None
+    total_completion_tokens_measured: Optional[int] = None
+    total_completion_tokens_estimated: Optional[int] = None
+    total_tokens_measured: Optional[int] = None
+    total_tokens_estimated: Optional[int] = None
+    tokens_display: Optional[str] = None
+    has_estimated_tokens: bool = False
     total_cost_usd: Optional[float] = None      # Sum of calculated costs or None if no pricing available
     failovers_count: int = 0
     failover_reasons: Dict[str, int] = field(default_factory=dict)
@@ -78,6 +103,7 @@ class TelemetryAggregates:
         return asdict(self)
 
 
+
 class TelemetryService:
     """Thread-safe persistent telemetry manager with bounded storage and honest aggregation."""
 
@@ -175,16 +201,42 @@ class TelemetryService:
         total_tokens: Optional[int] = None,
         failover_count: int = 0,
         error_category: Optional[str] = None,
+        prompt_tokens_measured: Optional[int] = None,
+        prompt_tokens_estimated: Optional[int] = None,
+        completion_tokens_measured: Optional[int] = None,
+        completion_tokens_estimated: Optional[int] = None,
+        total_tokens_measured: Optional[int] = None,
+        total_tokens_estimated: Optional[int] = None,
+        is_estimated: bool = False,
     ) -> TelemetryRecord:
         """Record an invocation attempt into memory and rotated log."""
         now = time.time()
         iso = datetime.datetime.fromtimestamp(now, datetime.timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ")
 
-        # Derive total tokens if prompt/completion available
-        if total_tokens is None and prompt_tokens is not None and completion_tokens is not None:
-            total_tokens = prompt_tokens + completion_tokens
+        if is_estimated:
+            p_est = prompt_tokens_estimated if prompt_tokens_estimated is not None else prompt_tokens
+            c_est = completion_tokens_estimated if completion_tokens_estimated is not None else completion_tokens
+            t_est = total_tokens_estimated if total_tokens_estimated is not None else total_tokens
+            if t_est is None and p_est is not None and c_est is not None:
+                t_est = p_est + c_est
+            p_meas, c_meas, t_meas = None, None, None
+            p_tok, c_tok, t_tok = p_est, c_est, t_est
+        else:
+            p_meas = prompt_tokens_measured if prompt_tokens_measured is not None else prompt_tokens
+            c_meas = completion_tokens_measured if completion_tokens_measured is not None else completion_tokens
+            t_meas = total_tokens_measured if total_tokens_measured is not None else total_tokens
+            if t_meas is None and p_meas is not None and c_meas is not None:
+                t_meas = p_meas + c_meas
+            p_est = prompt_tokens_estimated
+            c_est = completion_tokens_estimated
+            t_est = total_tokens_estimated
+            if t_est is None and p_est is not None and c_est is not None:
+                t_est = p_est + c_est
+            p_tok = p_meas if p_meas is not None else p_est
+            c_tok = c_meas if c_meas is not None else c_est
+            t_tok = t_meas if t_meas is not None else t_est
 
-        cost_usd = self.compute_cost(model, prompt_tokens, completion_tokens)
+        cost_usd = self.compute_cost(model, p_tok, c_tok)
 
         record = TelemetryRecord(
             timestamp=now,
@@ -195,9 +247,16 @@ class TelemetryService:
             model=model,
             outcome=outcome,
             latency_seconds=round(max(0.0, float(latency_seconds)), 4),
-            prompt_tokens=prompt_tokens,
-            completion_tokens=completion_tokens,
-            total_tokens=total_tokens,
+            prompt_tokens=p_tok,
+            completion_tokens=c_tok,
+            total_tokens=t_tok,
+            prompt_tokens_measured=p_meas,
+            prompt_tokens_estimated=p_est,
+            completion_tokens_measured=c_meas,
+            completion_tokens_estimated=c_est,
+            total_tokens_measured=t_meas,
+            total_tokens_estimated=t_est,
+            is_estimated=is_estimated or (p_est is not None and p_meas is None),
             cost_usd=cost_usd,
             failover_count=failover_count,
             error_category=error_category,
@@ -267,6 +326,13 @@ class TelemetryService:
                         prompt_tokens=d.get("prompt_tokens"),
                         completion_tokens=d.get("completion_tokens"),
                         total_tokens=d.get("total_tokens"),
+                        prompt_tokens_measured=d.get("prompt_tokens_measured"),
+                        prompt_tokens_estimated=d.get("prompt_tokens_estimated"),
+                        completion_tokens_measured=d.get("completion_tokens_measured"),
+                        completion_tokens_estimated=d.get("completion_tokens_estimated"),
+                        total_tokens_measured=d.get("total_tokens_measured"),
+                        total_tokens_estimated=d.get("total_tokens_estimated"),
+                        is_estimated=bool(d.get("is_estimated", False)),
                         cost_usd=d.get("cost_usd"),
                         failover_count=int(d.get("failover_count", 0)),
                         error_category=d.get("error_category"),
@@ -317,6 +383,14 @@ class TelemetryService:
                 total_prompt_tokens=None,
                 total_completion_tokens=None,
                 total_tokens=None,
+                total_prompt_tokens_measured=None,
+                total_prompt_tokens_estimated=None,
+                total_completion_tokens_measured=None,
+                total_completion_tokens_estimated=None,
+                total_tokens_measured=None,
+                total_tokens_estimated=None,
+                tokens_display=None,
+                has_estimated_tokens=False,
                 total_cost_usd=None,
                 failovers_count=0,
                 failover_reasons={},
@@ -332,6 +406,15 @@ class TelemetryService:
         prompt_tokens_sum = 0
         completion_tokens_sum = 0
         has_any_token_data = False
+        prompt_meas_sum = 0
+        prompt_est_sum = 0
+        has_meas_prompt = False
+        has_est_prompt = False
+        comp_meas_sum = 0
+        comp_est_sum = 0
+        has_meas_comp = False
+        has_est_comp = False
+        has_estimated_tokens = False
         costs_sum = 0.0
         has_any_cost_data = False
         failovers_count = 0
@@ -357,6 +440,41 @@ class TelemetryService:
                 completion_tokens_sum += r.completion_tokens
                 has_any_token_data = True
 
+            if r.prompt_tokens_measured is not None:
+                prompt_meas_sum += r.prompt_tokens_measured
+                has_meas_prompt = True
+            elif r.prompt_tokens is not None and not r.is_estimated:
+                prompt_meas_sum += r.prompt_tokens
+                has_meas_prompt = True
+
+            if r.prompt_tokens_estimated is not None:
+                prompt_est_sum += r.prompt_tokens_estimated
+                has_est_prompt = True
+                has_estimated_tokens = True
+            elif r.prompt_tokens is not None and r.is_estimated:
+                prompt_est_sum += r.prompt_tokens
+                has_est_prompt = True
+                has_estimated_tokens = True
+
+            if r.completion_tokens_measured is not None:
+                comp_meas_sum += r.completion_tokens_measured
+                has_meas_comp = True
+            elif r.completion_tokens is not None and not r.is_estimated:
+                comp_meas_sum += r.completion_tokens
+                has_meas_comp = True
+
+            if r.completion_tokens_estimated is not None:
+                comp_est_sum += r.completion_tokens_estimated
+                has_est_comp = True
+                has_estimated_tokens = True
+            elif r.completion_tokens is not None and r.is_estimated:
+                comp_est_sum += r.completion_tokens
+                has_est_comp = True
+                has_estimated_tokens = True
+
+            if r.is_estimated:
+                has_estimated_tokens = True
+
             if r.cost_usd is not None:
                 costs_sum += r.cost_usd
                 has_any_cost_data = True
@@ -368,6 +486,9 @@ class TelemetryService:
 
         error_rate = round(failed_calls / total_calls, 4) if total_calls > 0 else 0.0
         total_tokens_sum = (prompt_tokens_sum + completion_tokens_sum) if has_any_token_data else None
+        tot_meas_sum = (prompt_meas_sum + comp_meas_sum) if (has_meas_prompt or has_meas_comp) else None
+        tot_est_sum = (prompt_est_sum + comp_est_sum) if (has_est_prompt or has_est_comp) else None
+        tokens_display = format_token_count(tot_meas_sum, tot_est_sum)
 
         return TelemetryAggregates(
             window_seconds=window_seconds,
@@ -382,6 +503,14 @@ class TelemetryService:
             total_prompt_tokens=prompt_tokens_sum if has_any_token_data else None,
             total_completion_tokens=completion_tokens_sum if has_any_token_data else None,
             total_tokens=total_tokens_sum,
+            total_prompt_tokens_measured=prompt_meas_sum if has_meas_prompt else None,
+            total_prompt_tokens_estimated=prompt_est_sum if has_est_prompt else None,
+            total_completion_tokens_measured=comp_meas_sum if has_meas_comp else None,
+            total_completion_tokens_estimated=comp_est_sum if has_est_comp else None,
+            total_tokens_measured=tot_meas_sum,
+            total_tokens_estimated=tot_est_sum,
+            tokens_display=tokens_display,
+            has_estimated_tokens=has_estimated_tokens,
             total_cost_usd=round(costs_sum, 4) if has_any_cost_data else None,
             failovers_count=failovers_count,
             failover_reasons=dict(failover_reasons),
@@ -389,6 +518,7 @@ class TelemetryService:
             has_data=True,
         )
 
+
     def get_breakdown(
         self,
         window_seconds: Optional[int] = 86400,
@@ -399,7 +529,7 @@ class TelemetryService:
         global_aggs = self.get_aggregates(window_seconds=window_seconds)
 
         providers = set(known_providers or ["antigravity", "openai-codex", "opencode-go"])
-        roles = set(known_roles or ["orchestrator", "coder-primary", "coder-secondary", "reviewer", "research", "fast"])
+        roles = set(known_roles or RoleRegistry.get_role_ids())
 
         with self._lock:
             for r in self._buffer:

src/antigravity_provider/router/ui/add_account_wizard.py

@@ -22,25 +22,12 @@ import customtkinter as ctk
 
 from antigravity_provider.router.ui.theme import Theme
 from antigravity_provider.router.ui.components import HubButton, HubCard, HubEntry, HubModal
-from antigravity_provider.router.auto_assigner import AutoAssigner
+from antigravity_provider.router.auto_assigner import AutoAssigner, ensure_profile_in_routing
 from antigravity_provider.router.profile_manager import ProfileAuthManager
 from antigravity_provider.router.router_config import load_router_config
 from antigravity_provider.router.unified_health import EventLogService
 
 
-def ensure_profile_in_routing(profile_id: str) -> tuple[bool, str]:
-    """Keep existing chain rank or route a newly introduced profile slot."""
-    config = load_router_config()
-    assigned_role = next(
-        (role_id for role_id, policy in config.roles.items() if profile_id in policy.preferred_chain),
-        "",
-    )
-    if assigned_role:
-        return True, f"Профиль уже входит в цепочку '{assigned_role}'"
-    _display_name, role_code, tier = AutoAssigner.get_display_name_and_role(profile_id)
-    return AutoAssigner.assign_profile_to_role(profile_id, role_code, is_primary=tier == "primary")
-
-
 class AddAccountWizard(HubModal):
     """4-Step Add Account Wizard with OAuth / API Key support and Auto-Assignment."""
 

src/antigravity_provider/router/ui/routing_graph.py

@@ -127,7 +127,8 @@ def validate_graph(graph: RoutingGraph, config: Optional[RouterConfig] = None) -
     node_set = set(node_ids)
     for role_id in sorted({item for item in node_ids if node_ids.count(item) > 1}):
         issues.append(GraphIssue("duplicate-node", f"Роль {role_id} добавлена дважды", role_id))
-    if "orchestrator" not in node_set:
+    orch_node = next((n for n in ("manager", "orchestrator") if n in node_set), None)
+    if not orch_node:
         issues.append(GraphIssue("missing-orchestrator", "Отсутствует узел оркестратора"))
     for node in graph.nodes:
         policy = config.roles.get(node.role_id)
@@ -172,8 +173,8 @@ def validate_graph(graph: RoutingGraph, config: Optional[RouterConfig] = None) -
             visit(target)
         active.remove(role_id)
 
-    if "orchestrator" in node_set:
-        visit("orchestrator")
+    if orch_node:
+        visit(orch_node)
         for role_id in sorted(node_set - visited):
             issues.append(GraphIssue("unreachable", f"Роль {role_id} недостижима от оркестратора", role_id))
     return issues

src/antigravity_provider/router/unified_health.py

@@ -9,6 +9,7 @@ Single Source of Truth for:
 - EventLogService
 """
 from __future__ import annotations
+from antigravity_provider.router.role_registry import RoleRegistry
 
 import json
 import logging
@@ -644,21 +645,38 @@ class UnifiedHealthService:
         config = load_router_config()
         self.scan_all(force=False)
 
-        ROLE_META = {
-            "orchestrator": ("Главный оркестратор", "Управление командой, планирование, контроль исполнения"),
-            "coder-primary": ("Кодер 1", "Основная разработка кода и исправление дефектов"),
-            "coder-secondary": ("Кодер 2", "Параллельная разработка и вспомогательные модули"),
-            "reviewer": ("Ревьюер", "Независимое fail-closed ревью и валидация diff"),
-            "research": ("Исследователь", "Read-only поиск в кодовой базе и сбор фактов"),
-            "fast": ("Быстрый агент", "Оперативные вызовы, вспомогательные проверки"),
-            "universal": ("Универсальный агент", "Широкий спектр общих задач"),
-        }
-
         agents: List[AgentViewModel] = []
 
         for rname, rpol in config.roles.items():
-            rname_ru, rdesc_ru = ROLE_META.get(rname, (rname, ""))
+            rname_ru = RoleRegistry.get_role_name_ru(rname, rname)
+            rdesc_ru = RoleRegistry.get_role_description_ru(rname)
+            is_implemented = RoleRegistry.is_role_implemented(rname)
+            
             chain = rpol.preferred_chain
+            
+            if not is_implemented:
+                agents.append(AgentViewModel(
+                    role_id=rname,
+                    role_name_ru=rname_ru,
+                    role_description_ru=rdesc_ru or "Роль объявлена, исполнение не реализовано",
+                    assigned_profile_id=None,
+                    assigned_display_name=None,
+                    provider="N/A",
+                    provider_display_name="N/A",
+                    model="-",
+                    account_identity="Роль объявлена, исполнение не реализовано",
+                    routing_position="Отключено",
+                    status="unimplemented",
+                    status_label_ru="Не реализовано",
+                    is_active=False,
+                    is_main_orchestrator=False,
+                    cooldown_remaining_sec=0,
+                    session_id=None,
+                    active_quota_status="unavailable",
+                    active_quota_label="Не применяется (роль не активна)",
+                ))
+                continue
+
             if not chain:
                 continue
 
@@ -703,7 +721,7 @@ class UnifiedHealthService:
                     status=active_pvm.health_state,
                     status_label_ru=active_pvm.health_label_ru,
                     is_active=(active_pvm.health_state == STATUS_HEALTHY),
-                    is_main_orchestrator=(rname == "orchestrator"),
+                    is_main_orchestrator=(rname == "manager"),
                     cooldown_remaining_sec=active_pvm.cooldown_remaining_sec,
                     session_id=None,
                     active_quota_status=active_quota_st,
@@ -772,15 +790,7 @@ class UnifiedHealthService:
         self.scan_all(force=False)
         pipelines: Dict[str, RolePipeline] = {}
 
-        ROLE_NAMES = {
-            "orchestrator": "Главный оркестратор",
-            "coder-primary": "Кодер 1 (Primary)",
-            "coder-secondary": "Кодер 2 (Secondary)",
-            "reviewer": "Ревьюер",
-            "research": "Исследователь",
-            "fast": "Быстрый агент",
-            "universal": "Универсальный агент",
-        }
+        
 
         for rname, rpol in config.roles.items():
             nodes: List[PipelineNode] = []
@@ -834,7 +844,7 @@ class UnifiedHealthService:
 
             pipelines[rname] = RolePipeline(
                 role_id=rname,
-                role_name_ru=ROLE_NAMES.get(rname, rname),
+                role_name_ru=RoleRegistry.get_role_name_ru(rname, rname),
                 default_model=rpol.default_model or "auto",
                 max_failover=rpol.max_failover_attempts,
                 session_affinity=rpol.session_affinity_enabled,

src/antigravity_provider/router/web/server.py

@@ -125,8 +125,17 @@ def health_check():
         }
     }
 
-def sanitize_snapshot(snap_dict: Any) -> Any:
+def sanitize_snapshot(snap_dict: Any, email_masking_mode: Optional[str] = None) -> Any:
     import re
+    if email_masking_mode is None:
+        try:
+            from antigravity_provider.router.settings_service import get_hub_settings
+            email_masking_mode = get_hub_settings().get("email_masking_mode", "none")
+        except Exception:
+            email_masking_mode = "none"
+
+    mode = str(email_masking_mode or "none").strip().lower()
+
     secret_patterns = [
         re.compile(r'((?:access_token|refresh_token|api_key|token|password|secret|key)=)([^\s&,"]+)', re.IGNORECASE),
         re.compile(r'(sk-[a-zA-Z0-9_\-]{8,})'),
@@ -134,6 +143,23 @@ def sanitize_snapshot(snap_dict: Any) -> Any:
         re.compile(r'(Bearer\s+)([a-zA-Z0-9_\-\.]{8,})', re.IGNORECASE),
     ]
 
+    email_pattern = re.compile(r'\b([A-Za-z0-9._%+-]+)@([A-Za-z0-9.-]+\.[A-Za-z]{2,})\b')
+
+    def _mask_email_match(match: re.Match) -> str:
+        local_part = match.group(1)
+        domain_part = match.group(2)
+        if mode == "full":
+            return "***@***.***"
+        elif mode == "partial":
+            if len(local_part) > 2:
+                masked = f"{local_part[0]}***{local_part[-1]}"
+            elif local_part:
+                masked = f"{local_part[0]}***"
+            else:
+                masked = "***"
+            return f"{masked}@{domain_part}"
+        return match.group(0)
+
     def _mask_str(val: str) -> str:
         res = val
         for pat in secret_patterns:
@@ -141,6 +167,8 @@ def sanitize_snapshot(snap_dict: Any) -> Any:
                 res = pat.sub(r'\g<1>***', res)
             elif pat.groups == 1:
                 res = pat.sub(r'***', res)
+        if mode in ("partial", "full"):
+            res = email_pattern.sub(_mask_email_match, res)
         return res
 
     def _sanitize(node):
@@ -156,6 +184,7 @@ def sanitize_snapshot(snap_dict: Any) -> Any:
         return node
     return _sanitize(snap_dict)
 
+
 @app.get("/api/snapshot")
 def get_snapshot(authorized: bool = Depends(get_auth_token)):
     snapshot = HubStateStore.get().get_snapshot()

src/antigravity_provider/router/web/static/index.html

@@ -5,6 +5,7 @@
   <meta name="viewport" content="width=device-width, initial-scale=1.0">
   <title>Hermes Hub — Панель управления</title>
   <link rel="stylesheet" href="style.css">
+  <link rel="stylesheet" href="static/workflow.css">
   <link rel="icon" href="data:image/svg+xml,<svg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 100 100'><text y='.9em' font-size='90'>⚡</text></svg>">
 </head>
 <body>
@@ -92,52 +93,53 @@
       <div class="content-scroll">
         <!-- 1. OVERVIEW VIEW -->
         <section id="view-overview" class="view-pane active">
-          <div class="overview-grid">
-            <div class="kpi-card">
-              <div class="kpi-label">Состояние системы</div>
-              <div class="kpi-value text-healthy" id="kpi-system-readiness">—</div>
-              <div class="kpi-sub" id="kpi-readiness-summary">—</div>
+          <div class="workflow-toolbar">
+            <div class="workflow-mode" role="group" aria-label="Режим workflow">
+              <button class="workflow-mode-btn active" id="workflow-mode-live" data-mode="live">LIVE</button>
+              <button class="workflow-mode-btn" id="workflow-mode-edit" data-mode="edit">EDIT</button>
             </div>
-            <div class="kpi-card">
-              <div class="kpi-label">Подключено аккаунтов</div>
-              <div class="kpi-value text-accent" id="kpi-total-accounts">—</div>
-              <div class="kpi-sub" id="kpi-accounts-sub">—</div>
-            </div>
-            <div class="kpi-card">
-              <div class="kpi-label">Готовых ролей</div>
-              <div class="kpi-value text-healthy" id="kpi-ready-roles">—</div>
-              <div class="kpi-sub" id="kpi-roles-sub">—</div>
-            </div>
-            <div class="kpi-card">
-              <div class="kpi-label">Провайдеры ИИ</div>
-              <div class="kpi-value text-info" id="kpi-providers-count">—</div>
-              <div class="kpi-sub" id="kpi-providers-sub">5 поддерживаемых систем</div>
+            <div class="workflow-toolbar-actions">
+              <span class="workflow-save-state" id="workflow-save-state">Сохранено</span>
+              <button class="btn btn-secondary" id="btn-workflow-save">Сохранить workflow</button>
+              <button class="btn btn-primary" id="btn-agent-add">+ Добавить агента</button>
             </div>
           </div>
 
-          <div class="section-card">
-            <div class="section-card-header" style="display:flex; justify-content:space-between; align-items:center;">
-              <div>
-                <div class="section-card-title">Схема маршрутизации запросов</div>
-                <div class="section-card-subtitle">Распределение агентов и цепочки отказоустойчивости</div>
-              </div>
-              <button class="btn btn-secondary btn-sm" id="btn-auto-assign" title="Автоматическое распределение аккаунтов по ролям">
-                Авто-распределение
-              </button>
-            </div>
-            <div class="route-diagram-container" id="overview-route-diagram">
-              <!-- Rendered by app.js -->
-            </div>
+          <div class="workflow-kpis" aria-label="Оперативные показатели">
+            <article class="workflow-kpi"><span>Активные задачи</span><strong id="workflow-kpi-active">Загрузка…</strong><small id="workflow-kpi-active-reason">Получение снапшота</small></article>
+            <article class="workflow-kpi"><span>Агенты онлайн</span><strong id="workflow-kpi-online">Загрузка…</strong><small id="workflow-kpi-online-reason">Получение снапшота</small></article>
+            <article class="workflow-kpi"><span>Среднее время ответа</span><strong id="workflow-kpi-latency">Загрузка…</strong><small id="workflow-kpi-latency-reason">Получение телеметрии</small></article>
+            <article class="workflow-kpi"><span>Использование токенов</span><strong id="workflow-kpi-tokens">Загрузка…</strong><small id="workflow-kpi-tokens-reason">Получение телеметрии</small></article>
+            <article class="workflow-kpi"><span>Успешность задач</span><strong id="workflow-kpi-success">Загрузка…</strong><small id="workflow-kpi-success-reason">Получение телеметрии</small></article>
           </div>
 
-          <div class="section-card">
-            <div class="section-card-header">
-              <div class="section-card-title">Провайдеры и доступность runtime</div>
-              <div class="section-card-subtitle">Статус локальных адаптеров и обнаруженные модели</div>
+          <div class="workflow-main-layout">
+            <section class="workflow-board-card">
+              <header class="workflow-board-header">
+                <div><strong id="workflow-title">Workflow</strong> <span id="workflow-run-state" class="workflow-run-state">Загрузка…</span></div>
+                <label>Итерация: <strong id="workflow-iteration">Н/Д</strong> / <input id="workflow-max-iterations" type="number" min="1" max="100" value="5"></label>
+                <div class="workflow-zoom"><button id="workflow-zoom-out" title="Уменьшить">−</button><span id="workflow-zoom-value">100%</span><button id="workflow-zoom-in" title="Увеличить">+</button><button id="workflow-fit" title="Вписать граф">⛶</button></div>
+              </header>
+              <div class="workflow-canvas" id="workflow-canvas" tabindex="0" aria-label="Редактор графа workflow">
+                <svg id="workflow-edges" class="workflow-edges" aria-hidden="true"><defs><marker id="wf-arrow" markerWidth="8" markerHeight="8" refX="7" refY="4" orient="auto"><path d="M0,0 L8,4 L0,8 z"></path></marker></defs><g id="workflow-edge-layer"></g></svg>
+                <div id="workflow-node-layer" class="workflow-node-layer"></div>
+                <div id="workflow-empty" class="workflow-empty hidden"><strong>В workflow пока нет агентов</strong><span>Добавьте агента, затем переключитесь в EDIT и соедините узлы.</span></div>
+                <div class="workflow-minimap" id="workflow-minimap" aria-label="Мини-карта"></div>
               </div>
-            <div class="providers-grid" id="overview-providers-summary">
-              <!-- Rendered by app.js -->
+              <footer class="workflow-legend">
+                <span><i class="state-waiting"></i>Ожидает</span><span><i class="state-working"></i>Работает</span><span><i class="state-reviewing"></i>Проверяет</span><span><i class="state-error"></i>Ошибка</span><span><i class="state-completed"></i>Завершено</span>
+                <span class="edge-legend success">→ Успех</span><span class="edge-legend return">⇢ Возврат</span>
+              </footer>
+            </section>
+
+            <aside class="workflow-inspector" id="workflow-inspector">
+              <div class="workflow-inspector-empty"><strong>INSPECTOR</strong><span>Выберите агента на графе</span></div>
+            </aside>
           </div>
+
+          <div class="workflow-bottom-grid">
+            <section class="workflow-events"><header><strong>Последние события</strong><button id="workflow-events-all">Журнал событий →</button></header><div id="workflow-events-list" class="workflow-events-list"><p>Загрузка событий…</p></div></section>
+            <section class="workflow-run-panel"><header><strong>Управление LIVE</strong></header><textarea id="workflow-task" placeholder="Опишите реальную задачу для workflow"></textarea><div><button class="btn btn-primary" id="btn-workflow-start">Запустить workflow</button><button class="btn btn-secondary" id="btn-workflow-stop">Остановить</button></div><p id="workflow-run-error" class="workflow-run-error"></p></section>
           </div>
         </section>
 
@@ -177,15 +179,49 @@
         </section>
 
         <!-- 3. ROUTING VIEW (Main Control Center) -->
+        
           <section id="view-routing" class="view-pane">
-          <div class="view-header-note">
-            Главный центр управления маршрутизацией: перетаскивайте узлы для смены приоритета (Основной → Резерв 1 → Резерв 2), настраивайте рабочие модели и управляйте составом цепочек.
+            <div class="view-header-note" style="display:flex; justify-content: space-between; align-items: center;">
+              <div>
+                <i class="fa-solid fa-circle-info" style="color:var(--status-warning);"></i>
+                Перетащите аккаунт из правой панели в нужную роль и установите порядок (приоритет) использования.
+                <br>
+                Система будет использовать аккаунты сверху вниз по списку при выполнении задач с учётом квот и состояния.
               </div>
-          <div class="routing-pipelines-list" id="routing-pipelines-container">
+            </div>
+            
+            <div class="routing-grid" style="display: flex; gap: 24px; height: calc(100vh - 150px);">
+              <!-- LEFT COLUMN: Roles -->
+              <div class="routing-left-col" id="routing-roles-container" style="flex: 2; overflow-y: auto; padding-right: 8px;">
                 <!-- Rendered by app.js -->
               </div>
+              
+              <!-- RIGHT COLUMN: Available Accounts -->
+              <div class="routing-right-col" style="flex: 1; display: flex; flex-direction: column; border: 1px solid var(--border); border-radius: var(--radius-md); background-color: var(--surface-muted);">
+                <div style="padding: 16px; border-bottom: 1px solid var(--border-subtle);">
+                  <div style="display: flex; justify-content: space-between; margin-bottom: 12px;">
+                    <h3 style="font-family: var(--font-title); font-size: 14px; text-transform: uppercase;">Доступные аккаунты</h3>
+                    <span id="available-accounts-count" class="text-muted" style="font-size: 12px;">0 аккаунтов</span>
+                  </div>
+                  <div class="input-group">
+                    <input type="text" id="routing-account-search" placeholder="Поиск по аккаунтам, ролям или моделям..." style="width: 100%;">
+                  </div>
+                  <div style="display: flex; gap: 8px; margin-top: 8px;">
+                    <select id="routing-filter-provider" style="flex:1;"><option value="all">Все провайдеры</option></select>
+                    <select id="routing-filter-status" style="flex:1;"><option value="all">Все статусы</option></select>
+                  </div>
+                </div>
+                <div class="available-accounts-list" id="routing-available-container" style="flex: 1; overflow-y: auto; padding: 12px; display: flex; flex-direction: column; gap: 8px;">
+                  <!-- Rendered by app.js -->
+                </div>
+                <div style="padding: 16px; text-align: center; font-size: 12px; color: var(--text-muted); border-top: 1px dashed var(--border-subtle);">
+                  <i class="fa-solid fa-arrows-up-down-left-right"></i> Перетащите аккаунт в нужную роль или нажмите «+» для добавления
+                </div>
+              </div>
+            </div>
           </section>
 
+
         <!-- 4. ANALYTICS VIEW (P0-1 & P0-5) -->
         <section id="view-analytics" class="view-pane">
           <div class="view-header-note">
@@ -241,9 +277,23 @@
             <!-- Readiness state banner rendered by app.js -->
           </div>
 
+          <div class="section-card" id="health-preflight-card">
+            <div class="section-card-header" style="display:flex; justify-content:space-between; align-items:center;">
+              <div>
+                <div class="section-card-title">Проверка готовности (Preflight / Dependency Agent)</div>
+                <div class="section-card-subtitle">Zero-quota аудит окружения, зависимостей, прав доступа и локальных серверов</div>
+              </div>
+              <button class="btn btn-secondary btn-sm" id="btn-run-preflight">Запустить проверку готовности</button>
+            </div>
+            <div id="preflight-results-container" style="padding-top:12px;">
+              <!-- Rendered by app.js -->
+            </div>
+          </div>
+
           <div class="section-card">
             <div class="section-card-header">
               <div class="section-card-title">Ресурсы хост-системы</div>
+
               <div class="section-card-subtitle">Мониторинг CPU, памяти, диска и сетевых параметров</div>
             </div>
             <div class="host-resources-grid" id="health-host-resources">
@@ -366,6 +416,19 @@
                 </select>
               </div>
             </div>
+            <div class="setting-row">
+              <div class="setting-info">
+                <div class="setting-label">Маскирование почты (PII)</div>
+                <div class="setting-desc">Уровень маскирования адресов электронной почты в снапшотах и логах</div>
+              </div>
+              <div class="setting-control">
+                <select id="setting-email-masking-mode" class="select-filter">
+                  <option value="none" selected>Без маскирования (по умолчанию)</option>
+                  <option value="partial">Частичное (v***@gmail.com)</option>
+                  <option value="full">Полное (***@***.***)</option>
+                </select>
+              </div>
+            </div>
             <div class="setting-row">
               <div class="setting-info">
                 <div class="setting-label">Тема оформления</div>
@@ -505,5 +568,15 @@
 
   <!-- Application Logic -->
   <script src="app.js"></script>
+  <script src="static/workflow.js"></script>
+  <script>
+    const urlParams = new URLSearchParams(window.location.search);
+    const theme = urlParams.get('theme');
+    if (theme) {
+      document.body.setAttribute('data-theme', theme);
+      const sel = document.getElementById('theme-selector');
+      if (sel) sel.value = theme;
+    }
+  </script>
 </body>
 </html>

src/antigravity_provider/router/web/static/style.css

@@ -1,33 +1,7 @@
-/* Hermes Hub Web Client — Dark Theme & Cockpit Design System */
 
+/* Theme definitions according to A29 Design System "Крона" */
 :root {
-  --bg-base: #061916;
-  --bg-sidebar: #08221E;
-  --bg-header: #071B18;
-  --bg-statusbar: #061512;
-  --bg-modal-backdrop: rgba(2, 11, 9, 0.85);
-
-  --surface: #0B2520;
-  --surface-hover: #12342C;
-  --surface-active: #194535;
-  --surface-selected: #173B2E;
-  --surface-muted: #091E1A;
-
-  --border: #36513B;
-  --border-subtle: #203A2D;
-  --border-accent: #B78525;
-  --border-hover: #537456;
-
-  --text-primary: #F8F0DC;
-  --text-secondary: #D1C7AE;
-  --text-muted: #8FA395;
-  --text-accent: #E0B84E;
-
-  --accent: #C89A2B;
-  --accent-hover: #E0B84E;
-  --accent-pressed: #A9781E;
-  --accent-dim: rgba(200, 154, 43, 0.15);
-
+  /* Common variables */
   --status-healthy: #72C943;
   --status-warning: #E1A62B;
   --status-error: #E45C4F;
@@ -44,14 +18,107 @@
   --radius-md: 6px;
   --radius-lg: 8px;
 
-  --font-ui: "Segoe UI", -apple-system, BlinkMacSystemFont, Roboto, sans-serif;
-  --font-title: "Cinzel", "Segoe UI", serif, sans-serif;
+  --font-ui: 'Inter', "Segoe UI", -apple-system, sans-serif;
+  --font-title: "Cinzel", "Cinzel Decorative", "Segoe UI", serif;
   --font-mono: "Consolas", "Courier New", monospace;
 
   --header-height: 58px;
   --sidebar-width: 230px;
 }
 
+/* Default DARK theme */
+:root, body[data-theme="dark"] {
+  --bg-base: #101510;
+  --bg-sidebar: #0D120D;
+  --bg-header: #0D120D;
+  --bg-statusbar: #0D120D;
+  --bg-modal-backdrop: rgba(16, 21, 16, 0.85);
+
+  --surface: #1A2A1F;
+  --surface-hover: #2F4A36;
+  --surface-active: #3C5C44;
+  --surface-selected: #28402F;
+  --surface-muted: #152219;
+
+  --border: #36513B;
+  --border-subtle: #203A2D;
+  --border-accent: #CDAA64;
+  --border-hover: #537456;
+
+  --text-primary: #F7F1E3;
+  --text-secondary: #B0B8B2;
+  --text-muted: #8FA395;
+  --text-accent: #CDAA64;
+
+  --accent: #CDAA64;
+  --accent-hover: #E0B84E;
+  --accent-pressed: #A9781E;
+  --accent-dim: rgba(205, 170, 100, 0.15);
+}
+
+/* MEDIUM theme (cream cards on green canvas) */
+body[data-theme="medium"] {
+  --bg-base: #1A2A1F;
+  --bg-sidebar: #152219;
+  --bg-header: #152219;
+  --bg-statusbar: #152219;
+  --bg-modal-backdrop: rgba(26, 42, 31, 0.85);
+
+  --surface: #F7F1E3;
+  --surface-hover: #F0EAD6;
+  --surface-active: #E6DFCB;
+  --surface-selected: #EAE3CF;
+  --surface-muted: #2F4A36;
+
+  --border: #D1C7AE;
+  --border-subtle: #E0D7BF;
+  --border-accent: #CDAA64;
+  --border-hover: #B7A88D;
+
+  --text-primary: #101510;
+  --text-secondary: #2F4A36;
+  --text-muted: #537456;
+  --text-accent: #B78525;
+
+  --accent: #B78525;
+  --accent-hover: #CDAA64;
+  --accent-pressed: #9A6F1D;
+  --accent-dim: rgba(183, 133, 37, 0.15);
+}
+
+/* LIGHT theme */
+body[data-theme="light"] {
+  --bg-base: #F7F1E3;
+  --bg-sidebar: #F0EAD6;
+  --bg-header: #F0EAD6;
+  --bg-statusbar: #E6DFCB;
+  --bg-modal-backdrop: rgba(247, 241, 227, 0.85);
+
+  --surface: #FFFFFF;
+  --surface-hover: #FDFBF7;
+  --surface-active: #F4EEDF;
+  --surface-selected: #FAF7F0;
+  --surface-muted: #F0EAD6;
+
+  --border: #D1C7AE;
+  --border-subtle: #EAE3CF;
+  --border-accent: #CDAA64;
+  --border-hover: #C5BAA1;
+
+  --text-primary: #101510;
+  --text-secondary: #2F4A36;
+  --text-muted: #537456;
+  --text-accent: #B78525;
+
+  --accent: #B78525;
+  --accent-hover: #CDAA64;
+  --accent-pressed: #9A6F1D;
+  --accent-dim: rgba(183, 133, 37, 0.15);
+}
+/* Hermes Hub Web Client — Dark Theme & Cockpit Design System */
+
+
+
 * {
   box-sizing: border-box;
   margin: 0;
@@ -1559,20 +1626,114 @@ body {
 }
 
 /* ── Light Theme Override ── */
-body[data-theme="light"], body.theme-light {
-  --bg-base: #F4F6F8;
-  --bg-sidebar: #E9ECEF;
-  --bg-header: #FFFFFF;
-  --bg-statusbar: #DEE2E6;
-  --surface: #FFFFFF;
-  --surface-hover: #F1F3F5;
-  --surface-active: #E9ECEF;
-  --surface-selected: #E2E6EA;
-  --surface-muted: #F8F9FA;
-  --border: #CED4DA;
-  --border-subtle: #DEE2E6;
-  --border-accent: #B78525;
-  --text-primary: #212529;
-  --text-secondary: #495057;
-  --text-muted: #6C757D;
+
+
+/* Drag and Drop Styles */
+.draggable-item {
+  cursor: grab;
+  user-select: none;
+}
+.draggable-item:active {
+  cursor: grabbing;
+}
+.drop-zone {
+  min-height: 40px;
+  border: 1px dashed var(--border);
+  border-radius: var(--radius-sm);
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  font-size: 11px;
+  color: var(--text-muted);
+  transition: all 0.2s;
+  margin-top: 8px;
+}
+.drop-zone.drag-over {
+  border-color: var(--accent);
+  background-color: var(--accent-dim);
+  color: var(--accent);
+}
+.role-section {
+  background-color: var(--surface);
+  border: 1px solid var(--border);
+  border-radius: var(--radius-md);
+  margin-bottom: 16px;
+  display: flex;
+  flex-direction: column;
+}
+.role-header {
+  padding: 12px 16px;
+  border-bottom: 1px solid var(--border-subtle);
+  display: flex;
+  justify-content: space-between;
+  align-items: center;
+}
+.role-header h4 {
+  font-family: var(--font-title);
+  color: var(--text-primary);
+  font-size: 14px;
+  margin: 0;
+  display: flex;
+  align-items: center;
+  gap: 8px;
+}
+.role-badge {
+  font-size: 10px;
+  background-color: var(--accent-dim);
+  color: var(--accent);
+  padding: 2px 6px;
+  border-radius: 4px;
+  font-family: var(--font-ui);
+}
+.account-row {
+  display: grid;
+  grid-template-columns: 24px 20px 200px 120px 120px 150px 80px 80px 24px;
+  align-items: center;
+  gap: 8px;
+  padding: 8px 12px;
+  border-bottom: 1px solid var(--border-subtle);
+  background-color: var(--surface);
+  font-size: 12px;
+}
+.account-row:last-child {
+  border-bottom: none;
+}
+.account-row.drag-over {
+  border-top: 2px solid var(--accent);
+}
+.drag-handle {
+  color: var(--text-muted);
+  cursor: grab;
+  text-align: center;
+}
+.provider-logo-sm {
+  width: 16px;
+  height: 16px;
+  border-radius: 2px;
+}
+.available-account-card {
+  background-color: var(--surface);
+  border: 1px solid var(--border-subtle);
+  border-radius: var(--radius-sm);
+  padding: 10px;
+  display: flex;
+  align-items: center;
+  gap: 10px;
+  cursor: grab;
+}
+.available-account-card:hover {
+  border-color: var(--border);
+  background-color: var(--surface-hover);
+}
+.grid-header {
+  display: grid;
+  grid-template-columns: 24px 20px 200px 120px 120px 150px 80px 80px 24px;
+  align-items: center;
+  gap: 8px;
+  padding: 8px 12px;
+  font-size: 10px;
+  color: var(--text-muted);
+  text-transform: uppercase;
+  border-bottom: 1px solid var(--border-subtle);
+  background-color: var(--surface-muted);
 }

src/antigravity_provider/router/web/static/workflow.css

@@ -0,0 +1,89 @@
+/* A30 overview/workflow components. Theme colors come from the shared tokens. */
+#view-overview { padding: 0; min-width: 860px; }
+.workflow-toolbar { display:flex; align-items:center; justify-content:space-between; gap:12px; margin-bottom:10px; }
+.workflow-mode { display:flex; padding:3px; border:1px solid var(--border-subtle); border-radius:var(--radius-md); background:var(--surface-muted); }
+.workflow-mode-btn { min-width:78px; border:0; border-radius:var(--radius-sm); padding:7px 16px; color:var(--text-muted); background:transparent; cursor:pointer; font-weight:700; }
+.workflow-mode-btn.active { color:var(--text-primary); background:var(--surface-active); box-shadow:inset 0 0 0 1px var(--border-accent); }
+.workflow-mode-btn:first-child.active { color:var(--status-healthy); }
+.workflow-toolbar-actions { display:flex; gap:8px; align-items:center; }
+.workflow-save-state { color:var(--text-muted); font-size:11px; }
+.workflow-save-state.dirty { color:var(--status-warning); }
+.workflow-kpis { display:grid; grid-template-columns:repeat(5,minmax(130px,1fr)); gap:8px; margin-bottom:10px; }
+.workflow-kpi { padding:11px 13px; min-height:78px; border:1px solid var(--border-subtle); border-radius:var(--radius-md); background:linear-gradient(145deg,var(--surface),var(--surface-muted)); }
+.workflow-kpi span,.workflow-kpi small { display:block; color:var(--text-muted); }
+.workflow-kpi strong { display:block; margin:4px 0 2px; font-size:20px; font-weight:650; color:var(--text-primary); }
+.workflow-kpi small { font-size:10px; line-height:1.25; }
+.workflow-main-layout { display:grid; grid-template-columns:minmax(580px,1fr) 330px; gap:10px; min-height:520px; }
+.workflow-board-card,.workflow-inspector,.workflow-events,.workflow-run-panel { border:1px solid var(--border-subtle); border-radius:var(--radius-md); background:var(--surface); overflow:hidden; }
+.workflow-board-header { min-height:45px; padding:8px 12px; display:flex; gap:18px; align-items:center; justify-content:space-between; border-bottom:1px solid var(--border-subtle); color:var(--text-secondary); }
+.workflow-board-header strong { color:var(--text-accent); text-transform:uppercase; }
+.workflow-board-header label { margin-left:auto; font-size:11px; color:var(--text-muted); }
+.workflow-board-header input { width:42px; padding:3px; color:var(--text-primary); background:var(--surface-muted); border:1px solid var(--border); border-radius:3px; }
+.workflow-run-state { margin-left:8px; font-size:11px; color:var(--text-muted); }
+.workflow-run-state.running { color:var(--status-healthy); }
+.workflow-run-state.failed,.workflow-run-state.interrupted { color:var(--status-error); }
+.workflow-zoom { display:flex; align-items:center; border:1px solid var(--border-subtle); border-radius:var(--radius-sm); }
+.workflow-zoom button { width:28px; height:27px; border:0; border-left:1px solid var(--border-subtle); color:var(--text-accent); background:transparent; cursor:pointer; }
+.workflow-zoom span { min-width:44px; text-align:center; font-size:11px; }
+.workflow-canvas { position:relative; height:430px; overflow:hidden; background-color:var(--bg-base); background-image:radial-gradient(var(--border-subtle) 1px,transparent 1px); background-size:18px 18px; }
+.workflow-edges,.workflow-node-layer { position:absolute; inset:0; width:100%; height:100%; transform-origin:0 0; }
+.workflow-edges { overflow:visible; pointer-events:none; }
+.workflow-edges path { fill:none; stroke:var(--accent); stroke-width:1.6; marker-end:url(#wf-arrow); }
+.workflow-edges path.success,.workflow-edges path.review_passed { stroke:var(--status-healthy); }
+.workflow-edges path.review_failed,.workflow-edges path.error { stroke:var(--status-error); stroke-dasharray:7 5; }
+.workflow-edge-label { fill:var(--text-muted); font:9px var(--font-ui); text-transform:uppercase; }
+.workflow-edge-hit { fill:none; stroke:transparent; stroke-width:14; pointer-events:stroke; cursor:pointer; }
+.workflow-node { position:absolute; width:190px; min-height:92px; padding:11px 12px; border:1px solid var(--border-accent); border-radius:9px; background:linear-gradient(145deg,var(--surface),var(--surface-muted)); box-shadow:0 8px 22px rgba(0,0,0,.18); cursor:pointer; user-select:none; }
+.workflow-node.selected { outline:2px solid var(--accent); box-shadow:0 0 18px var(--accent-dim); }
+.workflow-node.working { border-color:var(--status-healthy); box-shadow:0 0 20px rgba(114,201,67,.24); }
+.workflow-node.error { border-color:var(--status-error); }
+.workflow-node.completed { border-color:var(--status-disabled); }
+.workflow-node h3 { margin:0 0 2px; font-size:13px; font-weight:650; color:var(--text-primary); }
+.workflow-node p { overflow:hidden; margin:1px 0; color:var(--text-muted); font-size:10px; white-space:nowrap; text-overflow:ellipsis; }
+.workflow-node-status { display:flex; align-items:center; gap:5px; margin-top:6px; font-size:10px; color:var(--text-secondary); }
+.workflow-node-status i,.workflow-legend i { width:7px; height:7px; border-radius:50%; background:var(--status-info); }
+.workflow-node.working .workflow-node-status i,.state-working { background:var(--status-healthy)!important; }
+.workflow-node.reviewing .workflow-node-status i,.state-reviewing { background:var(--status-warning)!important; }
+.workflow-node.error .workflow-node-status i,.state-error { background:var(--status-error)!important; }
+.workflow-node.completed .workflow-node-status i,.state-completed { background:var(--status-disabled)!important; }
+.workflow-port { display:none; position:absolute; top:37px; width:13px; height:22px; border:1px solid var(--accent); border-radius:8px; background:var(--surface-active); }
+.workflow-mode-edit .workflow-port { display:block; }
+.workflow-port.in { left:-8px; }
+.workflow-port.out { right:-8px; cursor:crosshair; }
+.workflow-legend { min-height:39px; padding:8px 12px; display:flex; align-items:center; gap:14px; flex-wrap:wrap; border-top:1px solid var(--border-subtle); color:var(--text-muted); font-size:10px; }
+.workflow-legend span { display:flex; align-items:center; gap:5px; }
+.edge-legend.success { color:var(--status-healthy); }.edge-legend.return { color:var(--status-error); }
+.workflow-minimap { position:absolute; left:10px; bottom:10px; width:110px; height:68px; border:1px solid var(--border); background:rgba(6,25,22,.82); pointer-events:none; }
+.workflow-minimap i { position:absolute; width:16px; height:9px; border:1px solid var(--accent); background:var(--surface-muted); }
+.workflow-empty { position:absolute; inset:0; display:flex; flex-direction:column; justify-content:center; align-items:center; gap:6px; color:var(--text-muted); }
+.workflow-inspector { min-width:0; padding:12px; overflow:auto; max-height:520px; }
+.workflow-inspector-empty { height:100%; display:flex; flex-direction:column; justify-content:center; align-items:center; gap:8px; color:var(--text-muted); }
+.workflow-inspector h2 { margin:0; color:var(--text-accent); font:600 15px var(--font-title); }
+.inspector-tabs { display:flex; gap:2px; margin:10px 0; border-bottom:1px solid var(--border-subtle); overflow-x:auto; }
+.inspector-tabs button { padding:7px 6px; border:0; border-bottom:2px solid transparent; color:var(--text-muted); background:transparent; font-size:10px; cursor:pointer; }
+.inspector-tabs button.active { color:var(--text-accent); border-color:var(--accent); }
+.inspector-field { display:block; margin:9px 0; color:var(--text-muted); font-size:10px; }
+.inspector-field input,.inspector-field select,.inspector-field textarea,.workflow-run-panel textarea,.workflow-dialog input,.workflow-dialog select,.workflow-dialog textarea { width:100%; margin-top:4px; padding:7px 8px; color:var(--text-primary); border:1px solid var(--border); border-radius:var(--radius-sm); background:var(--surface-muted); font:12px var(--font-ui); }
+.inspector-field textarea { min-height:58px; resize:vertical; }
+.inspector-value { padding:7px 0; color:var(--text-secondary); font-size:11px; overflow-wrap:anywhere; }
+.inspector-actions { display:flex; gap:7px; flex-wrap:wrap; margin-top:12px; }
+.agent-file-card { margin-top:12px; padding:10px; border:1px solid var(--border-subtle); border-radius:var(--radius-sm); }
+.agent-file-card code { display:block; margin:4px 0 8px; color:var(--text-secondary); overflow-wrap:anywhere; }
+.workflow-bottom-grid { display:grid; grid-template-columns:1.4fr 1fr; gap:10px; margin-top:10px; }
+.workflow-events,.workflow-run-panel { min-height:145px; padding:11px; }
+.workflow-events header,.workflow-run-panel header { display:flex; justify-content:space-between; margin-bottom:8px; color:var(--text-secondary); }
+.workflow-events header button { border:0; color:var(--text-accent); background:none; cursor:pointer; }
+.workflow-events-list { max-height:110px; overflow:auto; }
+.workflow-event { display:grid; grid-template-columns:72px 1fr auto; gap:8px; padding:5px 0; border-bottom:1px solid var(--border-subtle); color:var(--text-secondary); font-size:10px; }
+.workflow-event time,.workflow-event em { color:var(--text-muted); font-style:normal; }
+.workflow-event.error { color:var(--status-error); }
+.workflow-run-panel textarea { min-height:68px; resize:vertical; }
+.workflow-run-panel>div { display:flex; gap:8px; margin-top:8px; }
+.workflow-run-error { margin-top:6px; color:var(--status-error); font-size:10px; }
+.workflow-dialog-backdrop { position:fixed; inset:0; z-index:1200; display:flex; justify-content:center; align-items:center; padding:24px; background:var(--bg-modal-backdrop); }
+.workflow-dialog { width:min(580px,100%); max-height:88vh; overflow:auto; padding:18px; border:1px solid var(--border-accent); border-radius:var(--radius-lg); background:var(--surface); box-shadow:0 24px 70px rgba(0,0,0,.45); }
+.workflow-dialog h2 { margin-bottom:12px; color:var(--text-accent); }
+.workflow-dialog textarea.agent-file-editor { min-height:380px; font-family:var(--font-mono); }
+.workflow-dialog-actions { display:flex; justify-content:flex-end; gap:8px; margin-top:14px; }
+.hidden { display:none!important; }
+@media (max-width:1200px) { .workflow-kpis { grid-template-columns:repeat(3,1fr); }.workflow-main-layout { grid-template-columns:minmax(560px,1fr) 290px; } }

src/antigravity_provider/router/web/static/workflow.js

@@ -0,0 +1,497 @@
+/* Hermes Hub A30 workflow canvas — vanilla JS, no build step. */
+'use strict';
+
+const workflowUi = {
+  initialized: false,
+  mode: 'live',
+  selectedAgentId: null,
+  selectedTab: 'main',
+  scale: 1,
+  dirty: false,
+  draftEdges: [],
+  draftPositions: {},
+  connectingFrom: null,
+  drag: null,
+};
+
+function wfEscape(value) {
+  return String(value ?? '').replace(/[&<>'"]/g, (char) => ({
+    '&': '&amp;', '<': '&lt;', '>': '&gt;', "'": '&#39;', '"': '&quot;',
+  })[char]);
+}
+
+function wfValue(value, suffix = '') {
+  return value === null || value === undefined ? null : `${value}${suffix}`;
+}
+
+function wfUnavailable(elementId, reason) {
+  const value = document.getElementById(elementId);
+  const detail = document.getElementById(`${elementId}-reason`);
+  if (value) value.textContent = 'Н/Д';
+  if (detail) detail.textContent = `Н/Д: ${reason}`;
+}
+
+function initWorkflowOverview() {
+  if (workflowUi.initialized) return;
+  workflowUi.initialized = true;
+  document.getElementById('workflow-mode-live')?.addEventListener('click', () => setWorkflowMode('live'));
+  document.getElementById('workflow-mode-edit')?.addEventListener('click', () => setWorkflowMode('edit'));
+  document.getElementById('btn-workflow-save')?.addEventListener('click', saveWorkflowDraft);
+  document.getElementById('btn-agent-add')?.addEventListener('click', openAgentCreateDialog);
+  document.getElementById('btn-workflow-start')?.addEventListener('click', startWorkflowRun);
+  document.getElementById('btn-workflow-stop')?.addEventListener('click', () => executeAction('stop_workflow', {}));
+  document.getElementById('workflow-events-all')?.addEventListener('click', () => switchView('logs'));
+  document.getElementById('workflow-max-iterations')?.addEventListener('change', markWorkflowDirty);
+  document.getElementById('workflow-zoom-in')?.addEventListener('click', () => setWorkflowScale(workflowUi.scale + 0.1));
+  document.getElementById('workflow-zoom-out')?.addEventListener('click', () => setWorkflowScale(workflowUi.scale - 0.1));
+  document.getElementById('workflow-fit')?.addEventListener('click', fitWorkflowGraph);
+  const canvas = document.getElementById('workflow-canvas');
+  canvas?.addEventListener('mousemove', workflowPointerMove);
+  canvas?.addEventListener('mouseup', workflowPointerUp);
+  canvas?.addEventListener('mouseleave', workflowPointerCancel);
+  window.addEventListener('resize', drawWorkflowEdges);
+}
+
+function renderWorkflowOverview(snapshot) {
+  initWorkflowOverview();
+  const workflow = snapshot?.workflow;
+  if (!workflow) {
+    renderWorkflowLoading('Backend не вернул поле workflow');
+    return;
+  }
+  if (workflow.is_loading) {
+    renderWorkflowLoading('Workflow загружается');
+    return;
+  }
+  if (workflow.unavailable_reason) {
+    renderWorkflowLoading(workflow.unavailable_reason);
+    return;
+  }
+  if (!workflowUi.dirty && !workflowUi.drag && !workflowUi.connectingFrom) {
+    workflowUi.draftEdges = (workflow.definition?.edges || []).map((edge) => ({ ...edge }));
+    workflowUi.draftPositions = Object.fromEntries((workflow.agents || []).map((agent) => [
+      agent.id, { ...(agent.position || { x: 80, y: 80 }) },
+    ]));
+  }
+  renderWorkflowKpis(snapshot);
+  renderWorkflowHeader(workflow);
+  renderWorkflowNodes(workflow);
+  renderWorkflowInspector(snapshot, workflow);
+  renderWorkflowEvents(workflow.events || []);
+  requestAnimationFrame(drawWorkflowEdges);
+}
+
+function renderWorkflowLoading(reason) {
+  ['workflow-kpi-active', 'workflow-kpi-online', 'workflow-kpi-latency', 'workflow-kpi-tokens', 'workflow-kpi-success']
+    .forEach((id) => wfUnavailable(id, reason));
+  const empty = document.getElementById('workflow-empty');
+  if (empty) {
+    empty.classList.remove('hidden');
+    empty.innerHTML = `<strong>Workflow недоступен</strong><span>${wfEscape(reason)}</span>`;
+  }
+}
+
+function renderWorkflowKpis(snapshot) {
+  const metrics = snapshot.metrics || {};
+  const telemetry = metrics.telemetry || {};
+  const global = telemetry.global || {};
+  const workflow = snapshot.workflow || {};
+  const run = workflow.run || {};
+  const active = run.status === 'running' ? 1 : 0;
+  setWorkflowKpi('workflow-kpi-active', String(active), 'Источник: workflow.run.status');
+
+  const readiness = snapshot.readiness || {};
+  if (readiness.roles_ready_count === null || readiness.roles_ready_count === undefined || readiness.total_roles === undefined) {
+    wfUnavailable('workflow-kpi-online', 'readiness не содержит число готовых ролей');
+  } else {
+    setWorkflowKpi('workflow-kpi-online', `${readiness.roles_ready_count} / ${readiness.total_roles}`, 'Источник: readiness');
+  }
+  if (!global.total_calls) {
+    wfUnavailable('workflow-kpi-latency', 'за 24 часа нет измеренных вызовов');
+    wfUnavailable('workflow-kpi-success', 'за 24 часа нет завершённых вызовов');
+  } else {
+    const latency = wfValue(global.latency_p50_ms, ' мс');
+    latency ? setWorkflowKpi('workflow-kpi-latency', latency, 'Медиана p50, telemetry') : wfUnavailable('workflow-kpi-latency', 'провайдер не вернул задержку');
+    const success = global.successful_calls / global.total_calls * 100;
+    setWorkflowKpi('workflow-kpi-success', `${success.toFixed(1)}%`, `${global.successful_calls} из ${global.total_calls}, telemetry`);
+  }
+  if (global.total_tokens === null || global.total_tokens === undefined) {
+    wfUnavailable('workflow-kpi-tokens', 'провайдеры не вернули usage');
+  } else {
+    setWorkflowKpi('workflow-kpi-tokens', new Intl.NumberFormat('ru-RU').format(global.total_tokens), 'Источник: telemetry usage, 24 ч');
+  }
+}
+
+function setWorkflowKpi(id, value, reason) {
+  const node = document.getElementById(id);
+  const detail = document.getElementById(`${id}-reason`);
+  if (node) node.textContent = value;
+  if (detail) detail.textContent = reason;
+}
+
+function renderWorkflowHeader(workflow) {
+  const definition = workflow.definition || {};
+  const run = workflow.run || {};
+  document.getElementById('workflow-title').textContent = definition.name || 'Workflow без названия';
+  const state = document.getElementById('workflow-run-state');
+  state.textContent = workflowRunLabel(run.status);
+  state.className = `workflow-run-state ${wfEscape(run.status || 'idle')}`;
+  document.getElementById('workflow-iteration').textContent = run.iteration || 'Н/Д';
+  if (!workflowUi.dirty) document.getElementById('workflow-max-iterations').value = definition.max_iterations || 1;
+  document.getElementById('workflow-run-error').textContent = run.error || '';
+  document.getElementById('btn-workflow-start').disabled = run.status === 'running' || run.status === 'stopping';
+  document.getElementById('btn-workflow-stop').disabled = run.status !== 'running';
+}
+
+function workflowRunLabel(status) {
+  return ({ idle: 'Нет активного запуска', running: '● Запущен', stopping: 'Останавливается', completed: 'Завершён', failed: 'Ошибка', stopped: 'Остановлен', interrupted: 'Прерван перезапуском' })[status] || `Н/Д: неизвестный статус ${status || 'не указан'}`;
+}
+
+function renderWorkflowNodes(workflow) {
+  const layer = document.getElementById('workflow-node-layer');
+  const canvas = document.getElementById('workflow-canvas');
+  const agents = workflow.agents || [];
+  canvas.classList.toggle('workflow-mode-edit', workflowUi.mode === 'edit');
+  document.getElementById('workflow-empty').classList.toggle('hidden', agents.length > 0);
+  layer.innerHTML = agents.map((agent) => {
+    const pos = workflowUi.draftPositions[agent.id] || agent.position || { x: 80, y: 80 };
+    const cfg = agent.execution_config || {};
+    const assignment = cfg.unavailable_reason
+      ? `Н/Д: ${cfg.unavailable_reason}`
+      : [cfg.provider, cfg.model, cfg.account].filter(Boolean).join(' · ');
+    return `<article class="workflow-node ${wfEscape(agent.runtime_state)} ${workflowUi.selectedAgentId === agent.id ? 'selected' : ''}" data-agent-id="${wfEscape(agent.id)}" style="left:${Number(pos.x) || 0}px;top:${Number(pos.y) || 0}px">
+      <button class="workflow-port in" aria-label="Вход"></button><button class="workflow-port out" aria-label="Создать связь"></button>
+      <h3>${wfEscape(agent.name)}</h3><p>${wfEscape(agent.role)}</p><p>${wfEscape(agent.agent_file)}</p><p title="${wfEscape(assignment)}">${wfEscape(assignment)}</p>
+      <div class="workflow-node-status"><i></i><span>${wfEscape(runtimeStateLabel(agent.runtime_state))}</span></div>
+    </article>`;
+  }).join('');
+  layer.style.transform = `scale(${workflowUi.scale})`;
+  layer.querySelectorAll('.workflow-node').forEach((node) => {
+    node.addEventListener('click', () => selectWorkflowAgent(node.dataset.agentId));
+    node.addEventListener('mousedown', beginNodeDrag);
+    node.querySelector('.workflow-port.out')?.addEventListener('mousedown', beginConnection);
+    node.querySelector('.workflow-port.in')?.addEventListener('mouseup', finishConnection);
+  });
+  renderWorkflowMinimap(agents);
+}
+
+function runtimeStateLabel(state) {
+  return ({ waiting: 'Ожидает', working: 'Работает', reviewing: 'Проверяет', error: 'Ошибка', completed: 'Завершено', not_implemented: 'Исполнение не реализовано' })[state] || `Н/Д: ${state || 'статус не получен'}`;
+}
+
+function setWorkflowMode(mode) {
+  if (mode === 'edit' && currentSnapshot?.workflow?.run?.status === 'running') {
+    showToast('EDIT недоступен во время LIVE-выполнения', 'warning');
+    return;
+  }
+  workflowUi.mode = mode;
+  document.querySelectorAll('.workflow-mode-btn').forEach((button) => button.classList.toggle('active', button.dataset.mode === mode));
+  renderWorkflowOverview(currentSnapshot);
+}
+
+function selectWorkflowAgent(agentId) {
+  if (workflowUi.drag?.moved) return;
+  workflowUi.selectedAgentId = agentId;
+  renderWorkflowOverview(currentSnapshot);
+}
+
+function beginNodeDrag(event) {
+  if (workflowUi.mode !== 'edit' || event.target.classList.contains('workflow-port')) return;
+  const node = event.currentTarget;
+  const position = workflowUi.draftPositions[node.dataset.agentId] || { x: node.offsetLeft, y: node.offsetTop };
+  workflowUi.drag = { id: node.dataset.agentId, startX: event.clientX, startY: event.clientY, original: { ...position }, moved: false };
+  event.preventDefault();
+}
+
+function workflowPointerMove(event) {
+  if (workflowUi.drag) {
+    const drag = workflowUi.drag;
+    const dx = (event.clientX - drag.startX) / workflowUi.scale;
+    const dy = (event.clientY - drag.startY) / workflowUi.scale;
+    if (Math.abs(dx) + Math.abs(dy) > 3) drag.moved = true;
+    workflowUi.draftPositions[drag.id] = { x: Math.max(0, drag.original.x + dx), y: Math.max(0, drag.original.y + dy) };
+    const node = document.querySelector(`.workflow-node[data-agent-id="${CSS.escape(drag.id)}"]`);
+    if (node) {
+      node.style.left = `${workflowUi.draftPositions[drag.id].x}px`;
+      node.style.top = `${workflowUi.draftPositions[drag.id].y}px`;
+    }
+    drawWorkflowEdges();
+  }
+}
+
+function workflowPointerUp(event) {
+  if (workflowUi.drag) {
+    if (workflowUi.drag.moved) markWorkflowDirty();
+    workflowUi.drag = null;
+  }
+  if (workflowUi.connectingFrom && event.target === document.getElementById('workflow-canvas')) {
+    workflowUi.connectingFrom = null;
+    showToast('Создание связи отменено', 'info');
+  }
+}
+
+function workflowPointerCancel() {
+  if (workflowUi.drag) {
+    workflowUi.draftPositions[workflowUi.drag.id] = workflowUi.drag.original;
+    workflowUi.drag = null;
+    renderWorkflowOverview(currentSnapshot);
+  }
+  workflowUi.connectingFrom = null;
+}
+
+function beginConnection(event) {
+  if (workflowUi.mode !== 'edit') return;
+  workflowUi.connectingFrom = event.currentTarget.closest('.workflow-node').dataset.agentId;
+  event.stopPropagation();
+  event.preventDefault();
+  showToast('Выберите входной порт целевого агента', 'info');
+}
+
+function finishConnection(event) {
+  const target = event.currentTarget.closest('.workflow-node').dataset.agentId;
+  const source = workflowUi.connectingFrom;
+  workflowUi.connectingFrom = null;
+  event.stopPropagation();
+  if (!source || source === target) return;
+  openEdgeDialog({ id: `edge-${Date.now()}`, source, target, condition: 'SUCCESS', label: '' }, true);
+}
+
+function drawWorkflowEdges() {
+  const canvas = document.getElementById('workflow-canvas');
+  const svg = document.getElementById('workflow-edges');
+  const layer = document.getElementById('workflow-edge-layer');
+  if (!canvas || !svg || !layer) return;
+  svg.setAttribute('viewBox', `0 0 ${canvas.clientWidth} ${canvas.clientHeight}`);
+  const parts = [];
+  workflowUi.draftEdges.forEach((edge) => {
+    const source = document.querySelector(`.workflow-node[data-agent-id="${CSS.escape(edge.source)}"]`);
+    const target = document.querySelector(`.workflow-node[data-agent-id="${CSS.escape(edge.target)}"]`);
+    if (!source || !target) return;
+    const x1 = (source.offsetLeft + source.offsetWidth) * workflowUi.scale;
+    const y1 = (source.offsetTop + source.offsetHeight / 2) * workflowUi.scale;
+    const x2 = target.offsetLeft * workflowUi.scale;
+    const y2 = (target.offsetTop + target.offsetHeight / 2) * workflowUi.scale;
+    const bend = Math.max(45, Math.abs(x2 - x1) * .42);
+    const path = `M ${x1} ${y1} C ${x1 + bend} ${y1}, ${x2 - bend} ${y2}, ${x2} ${y2}`;
+    const klass = String(edge.condition || '').toLowerCase();
+    const labelX = (x1 + x2) / 2;
+    const labelY = (y1 + y2) / 2 - 5;
+    parts.push(`<path class="${wfEscape(klass)}" d="${path}"></path><path class="workflow-edge-hit" data-edge-id="${wfEscape(edge.id)}" d="${path}"></path><text class="workflow-edge-label" x="${labelX}" y="${labelY}">${wfEscape(edge.label || edge.condition)}</text>`);
+  });
+  layer.innerHTML = parts.join('');
+  layer.querySelectorAll('.workflow-edge-hit').forEach((path) => path.addEventListener('click', () => {
+    const edge = workflowUi.draftEdges.find((item) => item.id === path.dataset.edgeId);
+    if (workflowUi.mode === 'edit' && edge) openEdgeDialog(edge, false);
+  }));
+}
+
+function renderWorkflowMinimap(agents) {
+  const minimap = document.getElementById('workflow-minimap');
+  if (!minimap) return;
+  const maxX = Math.max(900, ...agents.map((agent) => (workflowUi.draftPositions[agent.id]?.x || 0) + 200));
+  const maxY = Math.max(500, ...agents.map((agent) => (workflowUi.draftPositions[agent.id]?.y || 0) + 100));
+  minimap.innerHTML = agents.map((agent) => {
+    const pos = workflowUi.draftPositions[agent.id] || { x: 0, y: 0 };
+    return `<i style="left:${pos.x / maxX * 92}px;top:${pos.y / maxY * 55}px"></i>`;
+  }).join('');
+}
+
+function setWorkflowScale(value) {
+  workflowUi.scale = Math.max(.5, Math.min(1.6, Math.round(value * 10) / 10));
+  document.getElementById('workflow-zoom-value').textContent = `${Math.round(workflowUi.scale * 100)}%`;
+  renderWorkflowOverview(currentSnapshot);
+}
+
+function fitWorkflowGraph() {
+  const agents = currentSnapshot?.workflow?.agents || [];
+  const maxX = Math.max(...agents.map((agent) => (workflowUi.draftPositions[agent.id]?.x || 0) + 210), 600);
+  const maxY = Math.max(...agents.map((agent) => (workflowUi.draftPositions[agent.id]?.y || 0) + 110), 400);
+  const canvas = document.getElementById('workflow-canvas');
+  setWorkflowScale(Math.min(canvas.clientWidth / maxX, canvas.clientHeight / maxY, 1));
+}
+
+function markWorkflowDirty() {
+  workflowUi.dirty = true;
+  const state = document.getElementById('workflow-save-state');
+  state.textContent = 'Есть несохранённые изменения';
+  state.classList.add('dirty');
+}
+
+async function saveWorkflowDraft() {
+  const definition = currentSnapshot?.workflow?.definition || {};
+  const result = await executeAction('save_workflow', {
+    id: definition.id,
+    name: definition.name,
+    start_agent_id: definition.start_agent_id,
+    escalation_agent_id: definition.escalation_agent_id,
+    max_iterations: Number(document.getElementById('workflow-max-iterations').value),
+    edges: workflowUi.draftEdges,
+    agents: Object.entries(workflowUi.draftPositions).map(([id, position]) => ({ id, position })),
+  });
+  if (result.ok) {
+    workflowUi.dirty = false;
+    const state = document.getElementById('workflow-save-state');
+    state.textContent = 'Сохранено';
+    state.classList.remove('dirty');
+  }
+}
+
+function renderWorkflowInspector(snapshot, workflow) {
+  const box = document.getElementById('workflow-inspector');
+  const agent = (workflow.agents || []).find((item) => item.id === workflowUi.selectedAgentId);
+  if (!agent) {
+    box.innerHTML = '<div class="workflow-inspector-empty"><strong>INSPECTOR</strong><span>Выберите агента на графе</span></div>';
+    return;
+  }
+  const cfg = agent.execution_config || {};
+  const tabs = [['main', 'Основное'], ['model', 'Модель'], ['instructions', 'Инструкции'], ['tools', 'Инструменты'], ['memory', 'Память'], ['history', 'История']];
+  box.innerHTML = `<h2>INSPECTOR: ${wfEscape(agent.name)}</h2><div class="inspector-tabs">${tabs.map(([id, label]) => `<button data-tab="${id}" class="${workflowUi.selectedTab === id ? 'active' : ''}">${label}</button>`).join('')}</div><div id="inspector-tab-content"></div>`;
+  box.querySelectorAll('.inspector-tabs button').forEach((button) => button.addEventListener('click', () => {
+    workflowUi.selectedTab = button.dataset.tab;
+    renderWorkflowInspector(snapshot, workflow);
+  }));
+  const content = document.getElementById('inspector-tab-content');
+  if (workflowUi.selectedTab === 'main') {
+    content.innerHTML = `<label class="inspector-field">Название<input id="agent-edit-name" value="${wfEscape(agent.name)}"></label><label class="inspector-field">Роль<div class="inspector-value">${wfEscape(agent.role)}</div></label><label class="inspector-field">Описание<textarea id="agent-edit-description">${wfEscape(agent.description)}</textarea></label><label class="inspector-field">Runtime status<div class="inspector-value">${wfEscape(runtimeStateLabel(agent.runtime_state))}</div></label><label class="inspector-field">Текущая задача<div class="inspector-value">${wfEscape(workflow.run?.current_agent_id === agent.id ? workflow.run.current_task : 'Н/Д: агент сейчас не выполняется')}</div></label><div class="inspector-actions"><button class="btn btn-primary btn-sm" id="agent-save-main">Сохранить</button><button class="btn btn-secondary btn-sm" id="agent-delete">Удалить</button></div>`;
+    document.getElementById('agent-save-main').onclick = () => executeAction('update_agent', { agent_id: agent.id, name: document.getElementById('agent-edit-name').value, description: document.getElementById('agent-edit-description').value });
+    document.getElementById('agent-delete').onclick = () => deleteWorkflowAgent(agent);
+  } else if (workflowUi.selectedTab === 'model') {
+    renderAgentModelTab(content, snapshot, agent);
+  } else if (workflowUi.selectedTab === 'instructions') {
+    content.innerHTML = `<div class="agent-file-card"><strong>Agent File</strong><code>${wfEscape(agent.agent_file)}</code><span>${agent.agent_file_exists ? 'Файл существует' : 'Н/Д: файл отсутствует'}</span><div class="inspector-actions"><button class="btn btn-secondary btn-sm" id="agent-file-open">Открыть в редакторе</button></div></div>`;
+    document.getElementById('agent-file-open').onclick = () => openAgentFileEditor(agent);
+  } else if (workflowUi.selectedTab === 'tools') {
+    content.innerHTML = `<label class="inspector-field">Инструменты, через запятую<input id="agent-tools" value="${wfEscape((agent.tools || []).join(', '))}"></label><button class="btn btn-primary btn-sm" id="agent-tools-save">Сохранить</button>`;
+    document.getElementById('agent-tools-save').onclick = () => executeAction('update_agent', { agent_id: agent.id, tools: document.getElementById('agent-tools').value.split(',').map((v) => v.trim()).filter(Boolean) });
+  } else if (workflowUi.selectedTab === 'memory') {
+    content.innerHTML = `<div class="inspector-value">${Object.keys(agent.memory_configuration || {}).length ? `<pre>${wfEscape(JSON.stringify(agent.memory_configuration, null, 2))}</pre>` : 'Н/Д: конфигурация памяти не задана'}</div>`;
+  } else {
+    const history = (workflow.events || []).filter((event) => event.agent_id === agent.id);
+    content.innerHTML = history.length ? history.slice(-20).reverse().map((event) => `<div class="workflow-event ${wfEscape(event.level)}"><time>${wfEscape(formatWorkflowTime(event.timestamp))}</time><span>${wfEscape(event.message)}</span><em>${event.duration_seconds == null ? '' : `${event.duration_seconds} с`}</em></div>`).join('') : '<div class="inspector-value">Н/Д: у агента ещё нет запусков</div>';
+  }
+}
+
+function renderAgentModelTab(content, snapshot, agent) {
+  const profiles = Object.values(snapshot.all_profiles || {}).filter((profile) => isConnectedProfile(profile));
+  const cfg = agent.execution_config || {};
+  const providers = [...new Set(profiles.map((profile) => profile.provider))];
+  content.innerHTML = `<label class="inspector-field">Провайдер<select id="agent-provider"><option value="">Не назначен</option>${providers.map((provider) => `<option value="${wfEscape(provider)}" ${provider === cfg.provider ? 'selected' : ''}>${wfEscape(provider)}</option>`).join('')}</select></label><label class="inspector-field">Аккаунт<select id="agent-account"></select></label><label class="inspector-field">Модель<select id="agent-model"></select></label><label class="inspector-field">Температура<input id="agent-temperature" type="number" step="0.1" min="0" max="2" value="${cfg.temperature ?? ''}" placeholder="Н/Д: не задана"></label><label class="inspector-field">Макс. токенов<input id="agent-max-tokens" type="number" min="1" value="${cfg.max_tokens ?? ''}" placeholder="Н/Д: не задано"></label><label class="inspector-field">Таймаут, с<input id="agent-timeout" type="number" min="1" value="${cfg.timeout || ''}"></label><button class="btn btn-primary btn-sm" id="agent-model-save">Изменить конфигурацию</button>`;
+  const providerSelect = document.getElementById('agent-provider');
+  const accountSelect = document.getElementById('agent-account');
+  const modelSelect = document.getElementById('agent-model');
+  const refreshAccounts = () => {
+    const matches = profiles.filter((profile) => profile.provider === providerSelect.value);
+    accountSelect.innerHTML = matches.length ? matches.map((profile) => `<option value="${wfEscape(profile.profile_id)}" ${profile.profile_id === cfg.account ? 'selected' : ''}>${wfEscape(profile.account_identity || profile.display_name || profile.profile_id)}</option>`).join('') : '<option value="">Н/Д: нет подключённых аккаунтов</option>';
+    refreshModels();
+  };
+  const refreshModels = () => {
+    const profile = profiles.find((item) => item.profile_id === accountSelect.value);
+    const models = profile?.preferred_models || Object.values(profile?.model_states || {}).map((state) => state.display_name).filter(Boolean);
+    modelSelect.innerHTML = models.length ? models.map((model) => `<option value="${wfEscape(model)}" ${model === cfg.model ? 'selected' : ''}>${wfEscape(model)}</option>`).join('') : '<option value="">Н/Д: модели не обнаружены</option>';
+  };
+  providerSelect.onchange = refreshAccounts;
+  accountSelect.onchange = refreshModels;
+  refreshAccounts();
+  document.getElementById('agent-model-save').onclick = () => executeAction('update_agent', {
+    agent_id: agent.id, provider: providerSelect.value, profile_id: accountSelect.value, model: modelSelect.value,
+    temperature: nullableNumber('agent-temperature'), max_tokens: nullableNumber('agent-max-tokens'), timeout: nullableNumber('agent-timeout'),
+  });
+}
+
+function nullableNumber(id) {
+  const value = document.getElementById(id).value.trim();
+  return value === '' ? null : Number(value);
+}
+
+async function deleteWorkflowAgent(agent) {
+  let result = await executeAction('delete_agent', { agent_id: agent.id });
+  if (result.ok && result.data?.confirmation_required) {
+    const refs = result.data.consequences?.workflow_edges || [];
+    if (!confirm(`Агент участвует в маршруте и/или графе. Будут удалены связи: ${refs.length ? refs.join(', ') : 'нет'}. Продолжить?`)) return;
+    result = await executeAction('delete_agent', { agent_id: agent.id, force: true });
+  }
+  if (result.ok && result.data?.deleted) workflowUi.selectedAgentId = null;
+}
+
+function openAgentCreateDialog() {
+  const profiles = Object.values(currentSnapshot?.all_profiles || {}).filter((profile) => isConnectedProfile(profile));
+  openWorkflowDialog('Добавить агента', `<label>Название<input id="new-agent-name" required></label><label>Роль (произвольный идентификатор)<input id="new-agent-role" placeholder="security-reviewer"></label><label>Описание<textarea id="new-agent-description"></textarea></label><label>Аккаунт<select id="new-agent-account"><option value="">Не назначать</option>${profiles.map((profile) => `<option value="${wfEscape(profile.profile_id)}">${wfEscape(profile.provider)} · ${wfEscape(profile.account_identity || profile.profile_id)}</option>`).join('')}</select></label><label>Agent File<input id="new-agent-file" placeholder="agents/имя.md"></label>`, async (close) => {
+    const account = document.getElementById('new-agent-account').value;
+    const profile = profiles.find((item) => item.profile_id === account);
+    const result = await executeAction('create_agent', {
+      name: document.getElementById('new-agent-name').value,
+      role: document.getElementById('new-agent-role').value,
+      description: document.getElementById('new-agent-description').value,
+      account,
+      model: profile?.preferred_models?.[0] || null,
+      agent_file: document.getElementById('new-agent-file').value,
+    });
+    if (result.ok) close();
+  });
+}
+
+async function openAgentFileEditor(agent) {
+  const result = await executeAction('read_agent_file', { agent_id: agent.id });
+  if (!result.ok) return;
+  const data = result.data || {};
+  if (!data.exists) {
+    showToast(`Н/Д: ${data.reason}`, 'warning');
+    return;
+  }
+  openWorkflowDialog(`Agent File — ${agent.name}`, `<label>Путь<input value="${wfEscape(data.path)}" readonly></label><label>Markdown<textarea id="agent-file-content" class="agent-file-editor">${wfEscape(data.content)}</textarea></label><p id="agent-file-unsaved">Нет изменений</p>`, async (close) => {
+    const saved = await executeAction('save_agent_file', { agent_id: agent.id, content: document.getElementById('agent-file-content').value });
+    if (saved.ok) close();
+  }, 'Сохранить');
+  const editor = document.getElementById('agent-file-content');
+  editor.addEventListener('input', () => { document.getElementById('agent-file-unsaved').textContent = 'Есть несохранённые изменения'; });
+}
+
+function openEdgeDialog(edge, isNew) {
+  openWorkflowDialog(isNew ? 'Новое ребро' : 'Редактор ребра', `<label>От<input value="${wfEscape(edge.source)}" readonly></label><label>К<input value="${wfEscape(edge.target)}" readonly></label><label>Условие<select id="edge-condition">${['SUCCESS', 'REVIEW_PASSED', 'REVIEW_FAILED', 'NEXT', 'ERROR', 'ALWAYS'].map((condition) => `<option ${condition === edge.condition ? 'selected' : ''}>${condition}</option>`).join('')}</select></label><label>Подпись<input id="edge-label" value="${wfEscape(edge.label || '')}"></label>${isNew ? '' : '<button class="btn btn-secondary btn-sm" id="edge-delete">Удалить ребро</button>'}`, (close) => {
+    const updated = { ...edge, condition: document.getElementById('edge-condition').value, label: document.getElementById('edge-label').value };
+    const index = workflowUi.draftEdges.findIndex((item) => item.id === edge.id);
+    if (index >= 0) workflowUi.draftEdges[index] = updated; else workflowUi.draftEdges.push(updated);
+    markWorkflowDirty(); close(); drawWorkflowEdges();
+  });
+  if (!isNew) document.getElementById('edge-delete').onclick = () => {
+    workflowUi.draftEdges = workflowUi.draftEdges.filter((item) => item.id !== edge.id);
+    markWorkflowDirty(); document.querySelector('.workflow-dialog-backdrop')?.remove(); drawWorkflowEdges();
+  };
+}
+
+function openWorkflowDialog(title, body, onSave, saveLabel = 'Применить') {
+  document.querySelector('.workflow-dialog-backdrop')?.remove();
+  const backdrop = document.createElement('div');
+  backdrop.className = 'workflow-dialog-backdrop';
+  backdrop.innerHTML = `<div class="workflow-dialog"><h2>${wfEscape(title)}</h2>${body}<div class="workflow-dialog-actions"><button class="btn btn-secondary" data-dialog-cancel>Отмена</button><button class="btn btn-primary" data-dialog-save>${wfEscape(saveLabel)}</button></div></div>`;
+  document.body.appendChild(backdrop);
+  const close = () => backdrop.remove();
+  backdrop.querySelector('[data-dialog-cancel]').onclick = close;
+  backdrop.querySelector('[data-dialog-save]').onclick = () => onSave(close);
+  backdrop.addEventListener('click', (event) => { if (event.target === backdrop) close(); });
+}
+
+async function startWorkflowRun() {
+  if (workflowUi.dirty) {
+    showToast('Сначала сохраните изменения графа', 'warning');
+    return;
+  }
+  const task = document.getElementById('workflow-task').value.trim();
+  await executeAction('start_workflow', { task });
+  setWorkflowMode('live');
+}
+
+function renderWorkflowEvents(events) {
+  const box = document.getElementById('workflow-events-list');
+  if (!events.length) {
+    box.innerHTML = '<p>Н/Д: workflow ещё не создавал событий</p>';
+    return;
+  }
+  box.innerHTML = events.slice(-20).reverse().map((event) => `<div class="workflow-event ${wfEscape(event.level)}"><time>${wfEscape(formatWorkflowTime(event.timestamp))}</time><span>${wfEscape(event.message)}${event.error ? ` — ${wfEscape(event.error)}` : ''}</span><em>${wfEscape(event.type)}</em></div>`).join('');
+}
+
+function formatWorkflowTime(value) {
+  if (!value) return 'Н/Д';
+  const parsed = new Date(value);
+  return Number.isNaN(parsed.getTime()) ? value : parsed.toLocaleTimeString('ru-RU');
+}

src/antigravity_provider/router/workflow_service.py

@@ -0,0 +1,785 @@
+"""Persistent agents, workflow graph and live execution for Hermes Hub.
+
+The router role registry remains the source of truth for logical agents and
+Provider -> Account -> Model assignment.  This module adds the pieces that do
+not fit the routing schema: Agent Files, editor layout, workflow transitions,
+execution checkpoints and a bounded event journal.
+"""
+from __future__ import annotations
+
+import json
+import re
+import threading
+import time
+import uuid
+from dataclasses import asdict, dataclass, field
+from pathlib import Path
+from typing import Any, Optional
+
+from antigravity_provider import paths
+from antigravity_provider.router.router_config import RolePolicy, load_router_config, save_router_config
+
+
+AGENT_STATES = {"waiting", "working", "reviewing", "error", "completed", "not_implemented"}
+EDGE_CONDITIONS = {"SUCCESS", "REVIEW_PASSED", "REVIEW_FAILED", "NEXT", "ERROR", "ALWAYS"}
+
+
+def _utc_timestamp() -> str:
+    import datetime
+
+    return datetime.datetime.now(datetime.timezone.utc).isoformat(timespec="seconds")
+
+
+def _slug(value: str) -> str:
+    result = re.sub(r"[^a-z0-9]+", "-", value.strip().lower()).strip("-")
+    return result or f"agent-{uuid.uuid4().hex[:8]}"
+
+
+def sanitize_run_data(node: Any) -> Any:
+    """Recursively strip or mask any credentials or secret keys from workflow run state."""
+    secret_key_substrings = [
+        "api_key", "token", "password", "secret", "jwt", "bearer",
+        "access_token", "refresh_token", "client_secret", "authorization",
+    ]
+    if isinstance(node, dict):
+        sanitized: dict[str, Any] = {}
+        for k, v in node.items():
+            k_lower = str(k).lower()
+            if any(s in k_lower for s in secret_key_substrings) and k_lower not in ("auth_status", "author", "auth_required"):
+                sanitized[k] = "***"
+            else:
+                sanitized[k] = sanitize_run_data(v)
+        return sanitized
+    elif isinstance(node, list):
+        return [sanitize_run_data(x) for x in node]
+    elif isinstance(node, str):
+        val = node
+        val = re.sub(r'Bearer\s+[a-zA-Z0-9_\-\.]{8,}', 'Bearer ***', val, flags=re.IGNORECASE)
+        val = re.sub(r'sk-[a-zA-Z0-9_\-]{8,}', 'sk-***', val)
+        val = re.sub(r'gho_[a-zA-Z0-9_\-]{8,}', 'gho_***', val)
+        val = re.sub(r'((?:access_token|refresh_token|api_key|token|password|secret|key)=)([^\s&,"]+)', r'\g<1>***', val, flags=re.IGNORECASE)
+        return val
+    return node
+
+
+def get_last_run_state(run_state_path: Optional[Path] = None) -> Optional[dict[str, Any]]:
+    """Return the last saved workflow run state from workflow_run_state.json, if any."""
+    p = run_state_path or paths.get_workflow_run_state_path()
+    if not p.is_file():
+        return None
+    try:
+        data = json.loads(p.read_text(encoding="utf-8"))
+        if isinstance(data, dict):
+            return sanitize_run_data(data)
+    except Exception:
+        pass
+    return None
+
+
+def _safe_agent_file(value: str, agent_id: str) -> tuple[Path, str]:
+    """Resolve an Agent File below HERMES_HOME/agents and reject traversal."""
+    root = paths.get_agent_files_dir().resolve()
+    candidate_name = Path(value or f"{agent_id}.md").name
+    if not candidate_name.lower().endswith(".md"):
+        candidate_name += ".md"
+    target = (root / candidate_name).resolve()
+    if target.parent != root:
+        raise ValueError("Agent File должен находиться в каталоге agents")
+    return target, f"agents/{candidate_name}"
+
+
+@dataclass
+class AgentDefinition:
+    id: str
+    name: str
+    role: str
+    description: str = ""
+    agent_file: str = ""
+    tools: list[str] = field(default_factory=list)
+    memory_configuration: dict[str, Any] = field(default_factory=dict)
+    execution_policy: dict[str, Any] = field(default_factory=dict)
+    timeout: int = 180
+    temperature: Optional[float] = None
+    max_tokens: Optional[int] = None
+    position: dict[str, float] = field(default_factory=lambda: {"x": 80.0, "y": 80.0})
+    metadata: dict[str, Any] = field(default_factory=dict)
+
+
+@dataclass
+class WorkflowEdge:
+    id: str
+    source: str
+    target: str
+    condition: str = "SUCCESS"
+    label: str = ""
+
+
+@dataclass
+class WorkflowDefinition:
+    id: str = "main"
+    name: str = "Основной workflow"
+    edges: list[WorkflowEdge] = field(default_factory=list)
+    max_iterations: int = 5
+    escalation_agent_id: Optional[str] = None
+    start_agent_id: Optional[str] = None
+
+
+@dataclass
+class WorkflowEvent:
+    timestamp: str
+    type: str
+    message: str
+    level: str = "info"
+    run_id: Optional[str] = None
+    agent_id: Optional[str] = None
+    iteration: Optional[int] = None
+    provider: Optional[str] = None
+    account: Optional[str] = None
+    model: Optional[str] = None
+    duration_seconds: Optional[float] = None
+    error: Optional[str] = None
+
+
+class WorkflowService:
+    _instance: Optional["WorkflowService"] = None
+    _instance_lock = threading.Lock()
+
+    def __init__(self, state_path: Optional[Path] = None, run_state_path: Optional[Path] = None) -> None:
+        self.state_path = state_path or paths.get_workflow_state_path()
+        self.run_state_path = run_state_path or paths.get_workflow_run_state_path()
+        self._lock = threading.RLock()
+        self._stop = threading.Event()
+        self._thread: Optional[threading.Thread] = None
+        self.agents: dict[str, AgentDefinition] = {}
+        self.workflow = WorkflowDefinition()
+        self.events: list[WorkflowEvent] = []
+        self.run: dict[str, Any] = self._idle_run()
+        self._completed_steps: list[dict[str, Any]] = []
+        self._load()
+        self._migrate_router_roles()
+
+    @classmethod
+    def get(cls) -> "WorkflowService":
+        if cls._instance is None:
+            with cls._instance_lock:
+                if cls._instance is None:
+                    cls._instance = cls()
+        return cls._instance
+
+    @staticmethod
+    def _idle_run() -> dict[str, Any]:
+        return {
+            "id": None,
+            "status": "idle",
+            "current_agent_id": None,
+            "current_task": None,
+            "iteration": 0,
+            "started_at": None,
+            "finished_at": None,
+            "elapsed_seconds": None,
+            "last_result": None,
+            "error": None,
+            "agent_states": {},
+        }
+
+    def _load(self) -> None:
+        # 1. Load run state from workflow_run_state.json
+        if self.run_state_path.is_file():
+            try:
+                run_state = json.loads(self.run_state_path.read_text(encoding="utf-8"))
+                if isinstance(run_state, dict):
+                    self._completed_steps = list(run_state.get("completed_steps", []))
+                    if run_state.get("status") in {"RUNNING", "running", "STOPPING", "stopping"}:
+                        run_state["status"] = "INTERRUPTED"
+                        run_state["interruption_reason"] = "Прогон был прерван перезапуском сервера или сбоем процесса"
+                        run_state["updated_at"] = _utc_timestamp()
+                        sanitized = sanitize_run_data(run_state)
+                        self.run_state_path.parent.mkdir(parents=True, exist_ok=True)
+                        temp = self.run_state_path.with_suffix(".tmp")
+                        temp.write_text(json.dumps(sanitized, ensure_ascii=False, indent=2), encoding="utf-8")
+                        temp.replace(self.run_state_path)
+            except Exception:
+                self._completed_steps = []
+
+        # 2. Load workflow definition and events from workflow_state.json
+        if not self.state_path.is_file():
+            return
+        try:
+            raw = json.loads(self.state_path.read_text(encoding="utf-8"))
+            self.agents = {
+                item["id"]: AgentDefinition(**item)
+                for item in raw.get("agents", [])
+                if isinstance(item, dict) and item.get("id")
+            }
+            wf = raw.get("workflow") or {}
+            edges = [WorkflowEdge(**edge) for edge in wf.pop("edges", []) if isinstance(edge, dict)]
+            self.workflow = WorkflowDefinition(edges=edges, **wf)
+            self.events = [WorkflowEvent(**event) for event in raw.get("events", [])[-200:]]
+            self.run = raw.get("run") or self._idle_run()
+            if self.run.get("status") in {"running", "stopping"}:
+                self.run["status"] = "interrupted"
+                self.run["error"] = "Выполнение прервано перезапуском Hermes Hub; checkpoint сохранён"
+                self._event("WORKFLOW_INTERRUPTED", self.run["error"], level="warning")
+        except (OSError, ValueError, TypeError):
+            self.agents = {}
+            self.workflow = WorkflowDefinition()
+            self.events = []
+            self.run = self._idle_run()
+
+    def _save_run_state(
+        self,
+        status: str,
+        step_index: int = 0,
+        current_agent: Optional[str] = None,
+        iteration: int = 1,
+        completed_step: Optional[dict[str, Any]] = None,
+        interruption_reason: Optional[str] = None,
+    ) -> None:
+        if completed_step:
+            self._completed_steps.append(sanitize_run_data(completed_step))
+
+        state_payload = {
+            "run_id": self.run.get("id"),
+            "status": status.upper(),
+            "started_at": self.run.get("started_at"),
+            "updated_at": _utc_timestamp(),
+            "current_step_index": step_index,
+            "current_agent_id": current_agent,
+            "iteration_count": iteration,
+            "completed_steps": list(self._completed_steps),
+            "interruption_reason": interruption_reason,
+        }
+        sanitized = sanitize_run_data(state_payload)
+        try:
+            self.run_state_path.parent.mkdir(parents=True, exist_ok=True)
+            temp = self.run_state_path.with_suffix(".tmp")
+            temp.write_text(json.dumps(sanitized, ensure_ascii=False, indent=2), encoding="utf-8")
+            temp.replace(self.run_state_path)
+        except Exception:
+            pass
+
+    def get_last_run_state(self) -> Optional[dict[str, Any]]:
+        """Return the last saved run state from workflow_run_state.json."""
+        return get_last_run_state(self.run_state_path)
+
+    def _save(self) -> None:
+        self.state_path.parent.mkdir(parents=True, exist_ok=True)
+        payload = {
+            "schema_version": 1,
+            "agents": [asdict(agent) for agent in self.agents.values()],
+            "workflow": asdict(self.workflow),
+            "events": [asdict(event) for event in self.events[-200:]],
+            "run": self.run,
+        }
+        temp = self.state_path.with_suffix(".tmp")
+        temp.write_text(json.dumps(payload, ensure_ascii=False, indent=2), encoding="utf-8")
+        temp.replace(self.state_path)
+
+    def _migrate_router_roles(self) -> None:
+        config = load_router_config()
+        changed = False
+        for index, (role_id, policy) in enumerate(config.roles.items()):
+            if role_id in self.agents:
+                continue
+            target, relative = _safe_agent_file("", role_id)
+            name = role_id.replace("-", " ").title()
+            description = ""
+            try:
+                from antigravity_provider.router.role_registry import get_role_definition
+
+                definition = get_role_definition(role_id)
+                name = getattr(definition, "name", None) or getattr(definition, "display_name", None) or name
+                description = getattr(definition, "description", "")
+            except (ImportError, AttributeError, TypeError):
+                pass
+            agent = AgentDefinition(
+                id=role_id,
+                name=name,
+                role=role_id,
+                description=description,
+                agent_file=relative,
+                position={"x": 70.0 + (index % 2) * 280.0, "y": 45.0 + (index // 2) * 125.0},
+            )
+            self.agents[role_id] = agent
+            self._ensure_file(target, agent)
+            changed = True
+        if changed:
+            self._save()
+
+    @staticmethod
+    def _ensure_file(target: Path, agent: AgentDefinition) -> None:
+        if not target.exists():
+            body = f"# {agent.name}\n\n## Роль\n\n{agent.role}\n\n## Назначение\n\n{agent.description or 'Инструкции ещё не заполнены.'}\n"
+            target.write_text(body, encoding="utf-8")
+
+    def _execution_config(self, agent: AgentDefinition) -> dict[str, Any]:
+        config = load_router_config()
+        policy = config.roles.get(agent.role)
+        profile_id = policy.preferred_chain[0] if policy and policy.preferred_chain else None
+        profile = config.profiles.get(profile_id) if profile_id else None
+        model = policy.default_model if policy else None
+        if not model and profile and profile.preferred_models:
+            model = profile.preferred_models[0]
+        return {
+            "provider": profile.provider if profile else None,
+            "account": profile_id,
+            "model": model,
+            "timeout": agent.timeout,
+            "temperature": agent.temperature,
+            "max_tokens": agent.max_tokens,
+            "unavailable_reason": None if profile else "Для роли не назначен доступный аккаунт",
+        }
+
+    def snapshot(self) -> dict[str, Any]:
+        with self._lock:
+            agents = []
+            for agent in self.agents.values():
+                item = asdict(agent)
+                target, relative = _safe_agent_file(agent.agent_file, agent.id)
+                item["agent_file"] = relative
+                item["agent_file_exists"] = target.is_file()
+                item["execution_config"] = self._execution_config(agent)
+                item["runtime_state"] = (self.run.get("agent_states") or {}).get(agent.id, "waiting")
+                agents.append(item)
+            return {
+                "agents": agents,
+                "definition": asdict(self.workflow),
+                "run": dict(self.run),
+                "events": [asdict(event) for event in self.events[-60:]],
+                "is_loading": False,
+            }
+
+    def read_agent_file(self, agent_id: str) -> dict[str, Any]:
+        with self._lock:
+            agent = self._require_agent(agent_id)
+            target, relative = _safe_agent_file(agent.agent_file, agent.id)
+            if not target.is_file():
+                return {"path": relative, "exists": False, "content": None, "reason": "Файл не найден на диске"}
+            return {"path": relative, "exists": True, "content": target.read_text(encoding="utf-8")}
+
+    def save_agent_file(self, agent_id: str, content: str) -> dict[str, Any]:
+        with self._lock:
+            agent = self._require_agent(agent_id)
+            target, relative = _safe_agent_file(agent.agent_file, agent.id)
+            temporary = target.with_suffix(".md.tmp")
+            temporary.write_text(str(content), encoding="utf-8")
+            temporary.replace(target)
+            self._event("AGENT_FILE_SAVED", f"Сохранён Agent File {relative}", agent_id=agent_id)
+            self._save()
+            return {"path": relative, "exists": True}
+
+    def create_agent(self, data: dict[str, Any]) -> AgentDefinition:
+        name = str(data.get("name") or "").strip()
+        role = _slug(str(data.get("role") or name))
+        agent_id = _slug(str(data.get("id") or role))
+        if not name:
+            raise ValueError("Укажите название агента")
+        with self._lock:
+            if agent_id in self.agents:
+                raise ValueError("Агент с таким идентификатором уже существует")
+            profile_id = str(data.get("account") or data.get("profile_id") or "").strip()
+            config = load_router_config()
+            if profile_id and profile_id not in config.profiles:
+                raise ValueError("Выбранный аккаунт отсутствует в маршрутизаторе")
+            config.roles[role] = RolePolicy(
+                role_name=role,
+                preferred_chain=[profile_id] if profile_id else [],
+                fallback_capabilities=list(data.get("fallback_capabilities") or [role]),
+                default_model=data.get("model") or None,
+            )
+            if not save_router_config(config):
+                raise OSError("Не удалось сохранить назначение агента")
+            target, relative = _safe_agent_file(str(data.get("agent_file") or ""), agent_id)
+            agent = AgentDefinition(
+                id=agent_id,
+                name=name,
+                role=role,
+                description=str(data.get("description") or ""),
+                agent_file=relative,
+                tools=[str(item) for item in data.get("tools", [])],
+                memory_configuration=dict(data.get("memory_configuration") or {}),
+                execution_policy=dict(data.get("execution_policy") or {}),
+                timeout=max(1, int(data.get("timeout") or 180)),
+                temperature=float(data["temperature"]) if data.get("temperature") is not None else None,
+                max_tokens=int(data["max_tokens"]) if data.get("max_tokens") is not None else None,
+                position=dict(data.get("position") or {"x": 80.0, "y": 80.0}),
+            )
+            self.agents[agent_id] = agent
+            copy_from = data.get("copy_from")
+            if copy_from:
+                source = self.read_agent_file(str(copy_from))
+                target.write_text(source.get("content") or "", encoding="utf-8")
+            else:
+                self._ensure_file(target, agent)
+            self._event("AGENT_CREATED", f"Создан агент «{name}»", agent_id=agent_id)
+            self._save()
+            return agent
+
+    def update_agent(self, agent_id: str, data: dict[str, Any]) -> AgentDefinition:
+        with self._lock:
+            agent = self._require_agent(agent_id)
+            config = load_router_config()
+            policy = config.roles.get(agent.role)
+            if not policy:
+                policy = RolePolicy(role_name=agent.role)
+                config.roles[agent.role] = policy
+            profile_id = str(data.get("account") or data.get("profile_id") or "").strip()
+            if profile_id:
+                profile = config.profiles.get(profile_id)
+                if not profile:
+                    raise ValueError("Выбранный аккаунт отсутствует в маршрутизаторе")
+                requested_provider = str(data.get("provider") or "").strip()
+                if requested_provider and profile.provider != requested_provider:
+                    raise ValueError("Аккаунт не принадлежит выбранному провайдеру")
+                policy.preferred_chain = [profile_id] + [item for item in policy.preferred_chain if item != profile_id]
+            if "model" in data:
+                model = str(data.get("model") or "").strip() or None
+                if model and profile_id:
+                    profile = config.profiles[profile_id]
+                    if model not in profile.preferred_models:
+                        raise ValueError("Модель не доступна выбранному аккаунту")
+                policy.default_model = model
+            if not save_router_config(config):
+                raise OSError("Не удалось сохранить назначение агента")
+            for attr in ("name", "description"):
+                if attr in data:
+                    setattr(agent, attr, str(data[attr]).strip())
+            for attr in ("tools", "memory_configuration", "execution_policy", "position"):
+                if attr in data:
+                    setattr(agent, attr, type(getattr(agent, attr))(data[attr]))
+            for attr in ("timeout", "max_tokens"):
+                if attr in data and data[attr] is not None:
+                    setattr(agent, attr, int(data[attr]))
+            if "temperature" in data:
+                agent.temperature = float(data["temperature"]) if data["temperature"] is not None else None
+            self._event("AGENT_UPDATED", f"Обновлён агент «{agent.name}»", agent_id=agent_id)
+            self._save()
+            return agent
+
+    def delete_agent(self, agent_id: str, force: bool = False) -> dict[str, Any]:
+        with self._lock:
+            agent = self._require_agent(agent_id)
+            edge_ids = [edge.id for edge in self.workflow.edges if edge.source == agent_id or edge.target == agent_id]
+            route_used = bool(load_router_config().roles.get(agent.role))
+            consequences = {"workflow_edges": edge_ids, "routing_role": agent.role if route_used else None}
+            if (edge_ids or route_used) and not force:
+                return {"deleted": False, "confirmation_required": True, "consequences": consequences}
+            config = load_router_config()
+            config.roles.pop(agent.role, None)
+            if not save_router_config(config):
+                raise OSError("Не удалось удалить роль из маршрутизатора")
+            self.workflow.edges = [edge for edge in self.workflow.edges if edge.id not in edge_ids]
+            self.agents.pop(agent_id)
+            self._event("AGENT_DELETED", f"Удалён агент «{agent.name}»", agent_id=agent_id, level="warning")
+            self._save()
+            return {"deleted": True, "consequences": consequences}
+
+    def save_workflow(self, data: dict[str, Any]) -> WorkflowDefinition:
+        with self._lock:
+            if self.run.get("status") in {"running", "stopping"}:
+                raise ValueError("Нельзя менять граф в режиме LIVE во время выполнения")
+            edges: list[WorkflowEdge] = []
+            seen: set[str] = set()
+            for raw in data.get("edges", []):
+                source, target = str(raw.get("source") or ""), str(raw.get("target") or "")
+                condition = str(raw.get("condition") or "SUCCESS").upper()
+                if source not in self.agents or target not in self.agents:
+                    raise ValueError("Ребро ссылается на отсутствующего агента")
+                if condition not in EDGE_CONDITIONS:
+                    raise ValueError(f"Неизвестное условие перехода: {condition}")
+                edge_id = str(raw.get("id") or f"edge-{uuid.uuid4().hex[:10]}")
+                if edge_id in seen:
+                    raise ValueError("Идентификаторы рёбер должны быть уникальны")
+                seen.add(edge_id)
+                edges.append(WorkflowEdge(edge_id, source, target, condition, str(raw.get("label") or "")))
+            max_iterations = int(data.get("max_iterations") or self.workflow.max_iterations)
+            if not 1 <= max_iterations <= 100:
+                raise ValueError("Предел итераций должен быть от 1 до 100")
+            for raw_agent in data.get("agents", []):
+                agent = self.agents.get(str(raw_agent.get("id") or ""))
+                pos = raw_agent.get("position")
+                if agent and isinstance(pos, dict):
+                    agent.position = {"x": float(pos.get("x", 0)), "y": float(pos.get("y", 0))}
+            self.workflow = WorkflowDefinition(
+                id=str(data.get("id") or self.workflow.id),
+                name=str(data.get("name") or self.workflow.name),
+                edges=edges,
+                max_iterations=max_iterations,
+                escalation_agent_id=data.get("escalation_agent_id") or None,
+                start_agent_id=data.get("start_agent_id") or None,
+            )
+            self._event("WORKFLOW_SAVED", f"Сохранён workflow «{self.workflow.name}»")
+            self._save()
+            return self.workflow
+
+    def start(self, task: str) -> dict[str, Any]:
+        task = str(task or "").strip()
+        if not task:
+            raise ValueError("Для запуска укажите реальную задачу")
+        with self._lock:
+            if self._thread and self._thread.is_alive():
+                raise ValueError("Workflow уже выполняется")
+            start_id = self.workflow.start_agent_id or (next(iter(self.agents), None))
+            if not start_id or start_id not in self.agents:
+                raise ValueError("В workflow нет стартового агента")
+            self._stop.clear()
+            self._completed_steps = []
+            self.run = self._idle_run()
+            self.run.update({
+                "id": uuid.uuid4().hex,
+                "status": "running",
+                "current_agent_id": start_id,
+                "current_task": task,
+                "iteration": 1,
+                "started_at": _utc_timestamp(),
+            })
+            self._event("WORKFLOW_STARTED", "Workflow запущен", run_id=self.run["id"], iteration=1)
+            self._save()
+            self._save_run_state("RUNNING", step_index=0, current_agent=start_id, iteration=1)
+            self._thread = threading.Thread(target=self._execute, name="HermesWorkflow", daemon=True)
+            self._thread.start()
+            return dict(self.run)
+
+    def stop(self) -> dict[str, Any]:
+        with self._lock:
+            if self.run.get("status") != "running":
+                raise ValueError("Нет выполняющегося workflow")
+            self.run["status"] = "stopping"
+            self._stop.set()
+            self._event("WORKFLOW_STOP_REQUESTED", "Запрошена остановка workflow", level="warning")
+            self._save()
+            self._save_run_state("STOPPED", step_index=len(self._completed_steps), current_agent=self.run.get("current_agent_id"), iteration=self.run.get("iteration", 1), interruption_reason="Остановлено пользователем")
+            return dict(self.run)
+
+    def _execute(self) -> None:
+        from antigravity_provider.router.router_engine import get_router_engine
+
+        started = time.monotonic()
+        context = str(self.run.get("current_task") or "")
+        current = str(self.run.get("current_agent_id") or "")
+        visited: dict[str, int] = {}
+        try:
+            engine = get_router_engine()
+            engine.reload_config()
+            while current and not self._stop.is_set():
+                with self._lock:
+                    agent = self._require_agent(current)
+                    visited[current] = visited.get(current, 0) + 1
+                    iteration = max(visited.values())
+                    self.run.update({"current_agent_id": current, "iteration": iteration})
+                    step_idx = len(self._completed_steps)
+                    if iteration > self.workflow.max_iterations:
+                        message = f"Достигнут предел итераций: {self.workflow.max_iterations}"
+                        self.run.update({"status": "failed", "error": message})
+                        self._event("WORKFLOW_MAX_ITERATIONS", message, level="error", agent_id=current, iteration=iteration)
+                        self._save_run_state("FAILED", step_index=step_idx, current_agent=current, iteration=iteration, interruption_reason=message)
+                        break
+                    file_data = self.read_agent_file(current)
+                    if not file_data["exists"]:
+                        raise FileNotFoundError(f"{file_data['path']}: {file_data['reason']}")
+                    self.run.setdefault("agent_states", {})[current] = (
+                        "reviewing" if "review" in agent.role.lower() else "working"
+                    )
+                    self._event("AGENT_STARTED", f"{agent.name} начал выполнение", agent_id=current, iteration=iteration)
+                    self._save()
+                    self._save_run_state("RUNNING", step_index=step_idx, current_agent=current, iteration=iteration)
+                request = {
+                    "model": self._execution_config(agent).get("model"),
+                    "messages": [
+                        {"role": "system", "content": file_data["content"]},
+                        {"role": "user", "content": context},
+                    ],
+                    "timeout": agent.timeout,
+                    "metadata": {"role": agent.role, "workflow_run_id": self.run["id"]},
+                }
+                if agent.temperature is not None:
+                    request["temperature"] = agent.temperature
+                if agent.max_tokens is not None:
+                    request["max_tokens"] = agent.max_tokens
+                step_started = time.monotonic()
+                response = engine.route_request(request, role=agent.role, session_id=self.run["id"])
+                duration = round(time.monotonic() - step_started, 3)
+                text = self._response_text(response)
+                status = self._result_status(response, text)
+                metadata = response.get("router_metadata", {}) if isinstance(response, dict) else {}
+                with self._lock:
+                    self.run["last_result"] = {"status": status, "content": text, "router_metadata": metadata}
+                    self.run.setdefault("agent_states", {})[current] = (
+                        "error" if status in {"ERROR", "REVIEW_FAILED"} else "completed"
+                    )
+                    self._event(
+                        "AGENT_COMPLETED",
+                        f"{agent.name}: {status}",
+                        level="success" if status not in {"ERROR", "REVIEW_FAILED"} else "warning",
+                        agent_id=current,
+                        iteration=iteration,
+                        provider=metadata.get("provider"),
+                        account=metadata.get("profile_id"),
+                        model=metadata.get("selected_model") or (metadata.get("selection_trace") or {}).get("selected_model"),
+                        duration_seconds=duration,
+                        error=text if status == "ERROR" else None,
+                    )
+                    step_summary = {
+                        "step_index": step_idx,
+                        "agent_id": current,
+                        "agent_name": agent.name,
+                        "iteration": iteration,
+                        "status": status,
+                        "duration_seconds": duration,
+                        "provider": metadata.get("provider"),
+                        "account": metadata.get("profile_id"),
+                        "model": metadata.get("selected_model") or (metadata.get("selection_trace") or {}).get("selected_model"),
+                        "error": text if status in {"ERROR", "REVIEW_FAILED"} else None,
+                        "timestamp": _utc_timestamp(),
+                    }
+                    edge = next(
+                        (item for item in self.workflow.edges if item.source == current and item.condition in {status, "ALWAYS"}),
+                        None,
+                    )
+                    if not edge and status not in {"ERROR", "REVIEW_FAILED"}:
+                        edge = next(
+                            (item for item in self.workflow.edges if item.source == current and item.condition in {"SUCCESS", "NEXT"}),
+                            None,
+                        )
+                    if not edge:
+                        self.run["status"] = "failed" if status in {"ERROR", "REVIEW_FAILED"} else "completed"
+                        if status == "ERROR":
+                            self.run["error"] = text or "Провайдер вернул ERROR без текста"
+                        self._event(
+                            "WORKFLOW_FAILED" if self.run["status"] == "failed" else "WORKFLOW_COMPLETED",
+                            f"Workflow завершён со статусом {status}",
+                            level="error" if self.run["status"] == "failed" else "success",
+                            agent_id=current,
+                            iteration=iteration,
+                            error=self.run.get("error") if self.run["status"] == "failed" else None,
+                        )
+                        if status == "ERROR":
+                            try:
+                                from antigravity_provider.router.unified_health import EventLogService
+
+                                EventLogService.get().log(
+                                    "workflow",
+                                    f"Ошибка агента «{agent.name}»",
+                                    details=self.run["error"],
+                                    level="error",
+                                )
+                            except Exception:
+                                pass
+                        self._save_run_state(
+                            "FAILED" if self.run["status"] == "failed" else "COMPLETED",
+                            step_index=step_idx + 1,
+                            current_agent=current,
+                            iteration=iteration,
+                            completed_step=step_summary,
+                            interruption_reason=self.run.get("error") if self.run["status"] == "failed" else None,
+                        )
+                        break
+                    self._event(
+                        "WORKFLOW_TRANSITION",
+                        f"Переход {current} → {edge.target}: {edge.condition}",
+                        agent_id=current,
+                        iteration=iteration,
+                    )
+                    context = json.dumps({
+                        "original_task": self.run["current_task"],
+                        "previous_agent": current,
+                        "structured_result": {"status": status, "content": text},
+                    }, ensure_ascii=False)
+                    current = edge.target
+                    self._save()
+                    self._save_run_state(
+                        "RUNNING",
+                        step_index=step_idx + 1,
+                        current_agent=current,
+                        iteration=iteration,
+                        completed_step=step_summary,
+                    )
+            with self._lock:
+                if self._stop.is_set():
+                    self.run.update({"status": "stopped", "error": "Остановлено пользователем"})
+                    self._event("WORKFLOW_STOPPED", "Workflow остановлен пользователем", level="warning")
+                    self._save_run_state("STOPPED", step_index=len(self._completed_steps), current_agent=current, iteration=iteration, interruption_reason="Остановлено пользователем")
+        except Exception as exc:
+            with self._lock:
+                self.run.update({"status": "failed", "error": str(exc)})
+                self._event("PROVIDER_ERROR", "Ошибка выполнения workflow", level="error", agent_id=current, error=str(exc))
+                try:
+                    from antigravity_provider.router.unified_health import EventLogService
+
+                    EventLogService.get().log("workflow", "Ошибка выполнения workflow", details=str(exc), level="error")
+                except Exception:
+                    pass
+                self._save_run_state("FAILED", step_index=len(self._completed_steps), current_agent=current, iteration=visited.get(current, 1), interruption_reason=str(exc))
+        finally:
+            with self._lock:
+                self.run["finished_at"] = _utc_timestamp()
+                self.run["elapsed_seconds"] = round(time.monotonic() - started, 3)
+                self.run["current_agent_id"] = current or self.run.get("current_agent_id")
+                self._save()
+
+
+    @staticmethod
+    def _response_text(response: Any) -> str:
+        if not isinstance(response, dict):
+            return str(response)
+        choices = response.get("choices")
+        if isinstance(choices, list) and choices:
+            message = choices[0].get("message", {})
+            return str(message.get("content") or choices[0].get("text") or "")
+        return str(response.get("content") or response.get("text") or response.get("output") or "")
+
+    @staticmethod
+    def _result_status(response: Any, text: str) -> str:
+        if isinstance(response, dict):
+            explicit = response.get("status") or response.get("structured_status")
+            if explicit and str(explicit).upper() in EDGE_CONDITIONS:
+                return str(explicit).upper()
+        for status in ("REVIEW_FAILED", "REVIEW_PASSED", "SUCCESS", "ERROR"):
+            if re.search(rf"\b{status}\b", text.upper()):
+                return status
+        return "SUCCESS"
+
+    def _event(self, event_type: str, message: str, **kwargs: Any) -> None:
+        self.events.append(WorkflowEvent(_utc_timestamp(), event_type, message, **kwargs))
+        self.events = self.events[-200:]
+
+    def _require_agent(self, agent_id: str) -> AgentDefinition:
+        agent = self.agents.get(agent_id)
+        if not agent:
+            raise ValueError("Агент не найден")
+        return agent
+
+
+def execute_workflow_action(action: str, data: dict[str, Any]) -> dict[str, Any]:
+    """Execute a workflow action through the shared ActionExecutor layer."""
+    service = WorkflowService.get()
+    if action == "create_agent":
+        result = asdict(service.create_agent(data))
+    elif action == "update_agent":
+        result = asdict(service.update_agent(str(data.get("agent_id") or ""), data))
+    elif action == "delete_agent":
+        result = service.delete_agent(str(data.get("agent_id") or ""), bool(data.get("force")))
+    elif action == "read_agent_file":
+        result = service.read_agent_file(str(data.get("agent_id") or ""))
+    elif action == "save_agent_file":
+        result = service.save_agent_file(str(data.get("agent_id") or ""), str(data.get("content") or ""))
+    elif action == "save_workflow":
+        result = asdict(service.save_workflow(data))
+    elif action == "start_workflow":
+        result = service.start(str(data.get("task") or ""))
+    elif action == "stop_workflow":
+        result = service.stop()
+    else:
+        raise ValueError("Неизвестное действие workflow")
+    if action in {"create_agent", "update_agent", "delete_agent"}:
+        try:
+            from antigravity_provider.router.state_store import HubStateStore
+
+            HubStateStore.get().refresh(force_scan=False)
+        except Exception:
+            pass
+    return {"ok": True, "message": "Выполнено", "data": result}
+
+
+# Alias for execution service compatibility
+WorkflowExecutionService = WorkflowService

tests/conftest.py

@@ -4,10 +4,12 @@ Enforces:
 1. Zero modification to real user credentials or router_profiles.yaml.
 2. Complete filesystem sandboxing in temporary directory via HERMES_HOME.
 3. Offline execution for default test runs (network / live require explicit -m markers).
+4. No accidental connects to local llama.cpp (8081/8082) or non-loopback hosts.
 """
 from __future__ import annotations
 
 import os
+import socket
 import sys
 from pathlib import Path
 import pytest
@@ -16,6 +18,9 @@ REPO_SRC = Path(__file__).resolve().parent.parent / "src"
 if str(REPO_SRC) not in sys.path or sys.path[0] != str(REPO_SRC):
     sys.path.insert(0, str(REPO_SRC))
 
+BLOCKED_INFERENCE_PORTS = {8081, 8082}
+LOOPBACK_HOSTS = {"127.0.0.1", "::1", "localhost", "0.0.0.0", "::"}
+
 
 @pytest.fixture(autouse=True)
 def isolate_hermes_environment(tmp_path, monkeypatch):
@@ -35,6 +40,62 @@ def isolate_hermes_environment(tmp_path, monkeypatch):
     yield temp_hermes
 
 
+def _socket_host_port(address) -> tuple[str | None, int | None]:
+    if isinstance(address, tuple) and len(address) >= 2:
+        host = address[0]
+        port = address[1]
+        if isinstance(host, bytes):
+            host = host.decode("utf-8", errors="replace")
+        try:
+            return str(host), int(port)
+        except (TypeError, ValueError):
+            return str(host), None
+    return None, None
+
+
+def _reject_hermetic_connect(address) -> None:
+    host, port = _socket_host_port(address)
+    if port in BLOCKED_INFERENCE_PORTS:
+        raise RuntimeError(
+            f"hermetic tests must not contact local inference at {host}:{port}; "
+            "mock check_local_servers / urllib.request.urlopen"
+        )
+    if host and host not in LOOPBACK_HOSTS and not host.startswith("127."):
+        raise RuntimeError(
+            f"hermetic tests must not open network connections to {host}:{port}; "
+            "mark the test live/network or mock the call"
+        )
+
+
+@pytest.fixture(autouse=True)
+def block_external_network_in_hermetic_tests(request, monkeypatch):
+    """Fail fast instead of hanging on llama.cpp or cloud APIs."""
+    if request.node.get_closest_marker("live") or request.node.get_closest_marker("network"):
+        yield
+        return
+
+    real_connect = socket.socket.connect
+    real_connect_ex = socket.socket.connect_ex
+    real_create_connection = socket.create_connection
+
+    def guarded_connect(self, address):
+        _reject_hermetic_connect(address)
+        return real_connect(self, address)
+
+    def guarded_connect_ex(self, address):
+        _reject_hermetic_connect(address)
+        return real_connect_ex(self, address)
+
+    def guarded_create_connection(address, *args, **kwargs):
+        _reject_hermetic_connect(address)
+        return real_create_connection(address, *args, **kwargs)
+
+    monkeypatch.setattr(socket.socket, "connect", guarded_connect)
+    monkeypatch.setattr(socket.socket, "connect_ex", guarded_connect_ex)
+    monkeypatch.setattr(socket, "create_connection", guarded_create_connection)
+    yield
+
+
 def pytest_configure(config):
     config.addinivalue_line("markers", "ui: mark test as requiring CustomTkinter / Tk graphical environment")
 

tests/test_a31_preflight_state_batching_pii.py

@@ -0,0 +1,435 @@
+"""Tests for Task A31: Preflight Dependency Agent, Workflow Run State, Local Concurrency & Context Window, PII Masking, and Cost Controller Honesty.
+"""
+from __future__ import annotations
+
+import json
+from pathlib import Path
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from antigravity_provider import paths
+from antigravity_provider.router.action_handler import ActionExecutor
+from antigravity_provider.router.adapters.local_adapter import LocalLLMAdapter
+from antigravity_provider.router.preflight_service import PreflightCheckService, PreflightItem, PreflightReport
+from antigravity_provider.router.role_registry import CANONICAL_ROLES, RoleRegistry
+from antigravity_provider.router.router_config import (
+    RouterConfig,
+    RouterProfileConfig,
+    get_default_router_config,
+    load_router_config,
+)
+from antigravity_provider.router.settings_service import (
+    DEFAULT_SETTINGS,
+    get_hub_settings,
+    invalidate_settings_cache,
+    save_hub_settings,
+)
+from antigravity_provider.router.telemetry_service import (
+    TelemetryAggregates,
+    TelemetryRecord,
+    TelemetryService,
+    format_token_count,
+)
+from antigravity_provider.router.web.server import sanitize_snapshot
+from antigravity_provider.router.workflow_service import (
+    AgentDefinition,
+    WorkflowDefinition,
+    WorkflowExecutionService,
+    WorkflowService,
+    get_last_run_state,
+    sanitize_run_data,
+)
+
+
+# ============================================================================
+# P0-1: Preflight Dependency Agent
+# ============================================================================
+
+
+def test_dependency_agent_role_registered():
+    """Verify 13th role 'dependency-agent' and its canonical aliases in RoleRegistry."""
+    assert len(CANONICAL_ROLES) == 13
+    assert "dependency-agent" in CANONICAL_ROLES
+
+    role_def = CANONICAL_ROLES["dependency-agent"]
+    assert role_def.role_id == "dependency-agent"
+    assert role_def.display_name_ru == "Проверяющий готовность"
+    assert role_def.short_name_ru == "Готовность"
+    assert role_def.is_implemented is True
+    assert "preflight" in role_def.capabilities
+
+    # Test alias resolution
+    aliases = [
+        "dependency-agent",
+        "dependency_agent",
+        "preflight",
+        "проверяющий готовность",
+        "агент зависимостей",
+        "готовность",
+        "dependency",
+    ]
+    for alias in aliases:
+        canonical = RoleRegistry.resolve_role_name(alias)
+        assert canonical == "dependency-agent", f"Alias '{alias}' resolved to '{canonical}'"
+
+
+def test_preflight_service_cli_and_environment():
+    """Verify CLI tools and environment checks with controlled discovery."""
+    service = PreflightCheckService.get()
+
+    with patch("antigravity_provider.router.preflight_service.shutil.which", return_value="/opt/agy"), \
+         patch("antigravity_provider.router.preflight_service.importlib.util.find_spec", return_value=object()):
+        cli_items = service.check_cli_dependencies()
+    assert len(cli_items) >= 3
+    ids = {item.check_id for item in cli_items}
+    assert "cli_agy" in ids
+    assert "pkg_fastapi" in ids
+    assert "pkg_uvicorn" in ids
+    assert all(item.status == "PASS" for item in cli_items)
+
+    env_items = service.check_system_environment()
+    assert len(env_items) >= 3
+    env_ids = {item.check_id for item in env_items}
+    assert "env_hermes_home" in env_ids
+    assert "env_config_writable" in env_ids
+    assert "env_logs_writable" in env_ids
+
+
+def _fake_preflight_items() -> list:
+    return [
+        PreflightItem(check_id="cli_agy", name="CLI", status="PASS", message="mocked"),
+        PreflightItem(check_id="env_hermes_home", name="HOME", status="PASS", message="mocked"),
+        PreflightItem(check_id="auth_local-1", name="AUTH", status="WARN", message="mocked"),
+        PreflightItem(check_id="local_srv_local-1", name="LLM", status="PASS", message="mocked"),
+    ]
+
+
+def test_preflight_service_run_all_and_action():
+    """Verify run_all_checks orchestration without probing this machine."""
+    service = PreflightCheckService.get()
+    cli, env, auth, local = (
+        [_fake_preflight_items()[0]],
+        [_fake_preflight_items()[1]],
+        [_fake_preflight_items()[2]],
+        [_fake_preflight_items()[3]],
+    )
+
+    with patch.object(service, "check_cli_dependencies", return_value=cli) as mock_cli, \
+         patch.object(service, "check_system_environment", return_value=env) as mock_env, \
+         patch.object(service, "check_auth_credentials", return_value=auth) as mock_auth, \
+         patch.object(service, "check_local_servers", return_value=local) as mock_local, \
+         patch("antigravity_provider.router.preflight_service.urllib.request.urlopen") as mock_urlopen:
+        report = service.run_all_checks()
+        action_res = ActionExecutor.execute("run_preflight", {})
+
+    mock_cli.assert_called()
+    mock_env.assert_called()
+    mock_auth.assert_called()
+    mock_local.assert_called()
+    mock_urlopen.assert_not_called()
+
+    assert isinstance(report, PreflightReport)
+    assert report.passed_count == 3
+    assert report.warn_count == 1
+    assert report.failed_count == 0
+    assert len(report.checks) == 4
+    report_dict = report.to_dict()
+    assert "success" in report_dict
+    assert "checks" in report_dict
+    assert isinstance(report_dict["checks"], list)
+
+    assert "ok" in action_res
+    assert "message" in action_res
+    assert "data" in action_res
+    assert "checks" in action_res["data"]
+    assert mock_urlopen.call_count == 0
+
+
+@pytest.mark.live
+def test_preflight_live_local_servers():
+    """Optional live probe of configured local servers. Not part of hermetic pytest."""
+    service = PreflightCheckService.get()
+    items = service.check_local_servers()
+    assert isinstance(items, list)
+    assert items
+
+
+# ============================================================================
+# P0-2: Workflow Run State Manager
+# ============================================================================
+
+
+def test_workflow_run_state_sanitization():
+    """Verify recursive secret stripping in workflow run state."""
+    raw_state = {
+        "run_id": "test-run-123",
+        "status": "RUNNING",
+        "api_key": "sk-1234567890abcdef",
+        "token": "gho_secrettoken123456",
+        "nested": {
+            "password": "supersecretpass",
+            "auth_status": "ok",
+            "message": "Bearer eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJzdWIiOiIxMjM0NTY3ODkwIn0",
+            "extra_url": "https://example.com/callback?access_token=secret12345&foo=bar",
+        },
+        "step_list": [
+            {"account": "acc-1", "client_secret": "my-client-secret-123"},
+            {"safe_field": "public_data"},
+        ],
+    }
+
+    sanitized = sanitize_run_data(raw_state)
+
+    assert sanitized["api_key"] == "***"
+    assert sanitized["token"] == "***"
+    assert sanitized["nested"]["password"] == "***"
+    assert sanitized["nested"]["auth_status"] == "ok"
+    assert "Bearer ***" in sanitized["nested"]["message"]
+    assert "access_token=***" in sanitized["nested"]["extra_url"]
+    assert sanitized["step_list"][0]["client_secret"] == "***"
+    assert sanitized["step_list"][1]["safe_field"] == "public_data"
+
+
+def test_workflow_run_state_interrupted_on_startup(tmp_path: Path):
+    """Verify that a RUNNING state in workflow_run_state.json transitions to INTERRUPTED on reload."""
+    state_file = tmp_path / "workflow_state.json"
+    run_state_file = tmp_path / "workflow_run_state.json"
+
+    # Pre-populate run state with RUNNING status
+    initial_run_state = {
+        "run_id": "run-crash-test",
+        "status": "RUNNING",
+        "started_at": "2026-08-26T00:00:00Z",
+        "updated_at": "2026-08-26T00:00:00Z",
+        "current_step_index": 2,
+        "current_agent_id": "developer-1",
+        "iteration_count": 1,
+        "completed_steps": [
+            {"step_index": 0, "agent_id": "manager", "status": "SUCCESS"},
+            {"step_index": 1, "agent_id": "developer-1", "status": "WORKING"},
+        ],
+        "interruption_reason": None,
+    }
+    run_state_file.write_text(json.dumps(initial_run_state), encoding="utf-8")
+
+    # Initialize WorkflowService
+    service = WorkflowService(state_path=state_file, run_state_path=run_state_file)
+
+    # Check that state transitioned to INTERRUPTED
+    last_state = service.get_last_run_state()
+    assert last_state is not None
+    assert last_state["status"] == "INTERRUPTED"
+    assert last_state["interruption_reason"] == "Прогон был прерван перезапуском сервера или сбоем процесса"
+    assert len(last_state["completed_steps"]) == 2
+
+    # Verify top-level function
+    assert get_last_run_state(run_state_file)["status"] == "INTERRUPTED"
+
+
+def test_workflow_execution_service_alias():
+    """Verify WorkflowExecutionService is an alias of WorkflowService."""
+    assert WorkflowExecutionService is WorkflowService
+
+
+# ============================================================================
+# P0-3: Local Concurrency & Context Window
+# ============================================================================
+
+
+def test_local_profile_max_concurrency_is_one():
+    """Verify all local provider profiles have max_concurrency = 1."""
+    config = get_default_router_config()
+    for pid, pcfg in config.profiles.items():
+        if pcfg.provider == "local":
+            assert pcfg.max_concurrency == 1, f"Local profile {pid} has max_concurrency={pcfg.max_concurrency}"
+
+    # Verify loaded config also enforces max_concurrency = 1 for local profiles
+    loaded = load_router_config()
+    for pid, pcfg in loaded.profiles.items():
+        if pcfg.provider == "local":
+            assert pcfg.max_concurrency == 1
+
+
+def test_local_adapter_get_context_window():
+    """Verify LocalLLMAdapter retrieves context window accurately without hallucinating defaults."""
+    adapter = LocalLLMAdapter()
+
+    # Profile with explicit context_window in auth_config
+    prof_with_cfg = RouterProfileConfig(
+        profile_id="local-test-1",
+        provider="local",
+        account_id="acc-1",
+        auth_config={"context_window": 8192},
+    )
+    assert adapter.get_context_window(prof_with_cfg) == 8192
+
+    # Profile without context length and with non-responding server
+    prof_empty = RouterProfileConfig(
+        profile_id="local-test-2",
+        provider="local",
+        account_id="acc-2",
+        custom_base_url="http://127.0.0.1:9999/v1",
+    )
+    # Must return None instead of inventing fake numbers
+    assert adapter.get_context_window(prof_empty) is None
+
+
+def test_local_adapter_context_truncation_guard():
+    """Verify context truncation guard protects against VRAM overflow when context_window is known."""
+    adapter = LocalLLMAdapter()
+
+    prof = RouterProfileConfig(
+        profile_id="local-small-ctx",
+        provider="local",
+        account_id="acc-1",
+        auth_config={"context_window": 500},
+        custom_base_url="http://127.0.0.1:12345/v1",
+    )
+
+    # Huge prompt exceeding 500 tokens
+    long_middle_content = "important historical dialogue step " * 100
+    messages = [
+        {"role": "system", "content": "You are a helpful assistant."},
+        {"role": "user", "content": "Initial prompt 1"},
+        {"role": "assistant", "content": long_middle_content},
+        {"role": "user", "content": "Initial prompt 2"},
+        {"role": "assistant", "content": long_middle_content},
+        {"role": "user", "content": "Latest user task to execute."},
+    ]
+
+    mock_resp = {
+        "choices": [{"message": {"role": "assistant", "content": "Truncated prompt executed successfully."}}],
+        "usage": {"prompt_tokens": 200, "completion_tokens": 10},
+    }
+
+    with patch("urllib.request.urlopen") as mock_urlopen:
+        mock_cm = MagicMock()
+        mock_cm.read.return_value = json.dumps(mock_resp).encode("utf-8")
+        mock_urlopen.return_value.__enter__.return_value = mock_cm
+
+        res = adapter.invoke(prof, {"messages": messages, "max_tokens": 100})
+        assert res["choices"][0]["message"]["content"] == "Truncated prompt executed successfully."
+
+        # Verify sent payload messages were truncated
+        args, kwargs = mock_urlopen.call_args
+        sent_req = args[0]
+        sent_body = json.loads(sent_req.data.decode("utf-8"))
+        sent_messages = sent_body["messages"]
+
+        assert sent_messages[0]["role"] == "system"
+        assert sent_messages[-1]["content"] == "Latest user task to execute."
+        # Total count of messages should be pruned
+        assert len(sent_messages) < len(messages)
+
+
+# ============================================================================
+# P0-4: PII Email Masking
+# ============================================================================
+
+
+def test_settings_email_masking_mode():
+    """Verify email_masking_mode in default settings and persistence."""
+    assert DEFAULT_SETTINGS["email_masking_mode"] == "none"
+
+    settings = get_hub_settings()
+    assert settings.get("email_masking_mode") in ("none", "partial", "full")
+
+
+def test_sanitize_snapshot_email_masking_modes():
+    """Verify email masking behavior across 'none', 'partial', and 'full' modes."""
+    snapshot_data = {
+        "user_email": "vasya.pupkin@example.com",
+        "account_id": "google-user-1",
+        "api_key": "sk-secret123456789",
+        "nested": {
+            "developer": "developer.one@domain.org",
+            "reviewer": "r@test.com",
+        },
+    }
+
+    # 1. Mode: none (emails unchanged, secrets masked)
+    san_none = sanitize_snapshot(snapshot_data, email_masking_mode="none")
+    assert san_none["user_email"] == "vasya.pupkin@example.com"
+    assert san_none["nested"]["developer"] == "developer.one@domain.org"
+    assert san_none["nested"]["reviewer"] == "r@test.com"
+    assert "api_key" not in san_none
+
+    # 2. Mode: partial (preserves first and last char of local part + domain for differentiation)
+    san_partial = sanitize_snapshot(snapshot_data, email_masking_mode="partial")
+    assert san_partial["user_email"] == "v***n@example.com"
+    assert san_partial["nested"]["developer"] == "d***e@domain.org"
+    assert san_partial["nested"]["reviewer"] == "r***@test.com"
+    assert "api_key" not in san_partial
+
+    # 3. Mode: full (***@***.***)
+    san_full = sanitize_snapshot(snapshot_data, email_masking_mode="full")
+    assert san_full["user_email"] == "***@***.***"
+    assert san_full["nested"]["developer"] == "***@***.***"
+    assert san_full["nested"]["reviewer"] == "***@***.***"
+    assert "api_key" not in san_full
+
+
+# ============================================================================
+# P0-5: Cost Controller Token Honesty
+# ============================================================================
+
+
+def test_telemetry_measured_vs_estimated_tokens(tmp_path: Path):
+    """Verify telemetry distinguishes measured exact tokens from estimated tokens with ~."""
+    log_file = tmp_path / "telemetry_test.jsonl"
+    service = TelemetryService(log_path=log_file)
+
+    # 1. Record measured call
+    rec1 = service.record_call(
+        role="developer-1",
+        profile_id="ag-w1",
+        provider="antigravity",
+        model="claude-3-7-sonnet",
+        outcome="success",
+        latency_seconds=1.25,
+        prompt_tokens_measured=500,
+        completion_tokens_measured=150,
+        is_estimated=False,
+    )
+    assert rec1.prompt_tokens_measured == 500
+    assert rec1.prompt_tokens_estimated is None
+    assert rec1.is_estimated is False
+    assert rec1.total_tokens == 650
+
+    # 2. Record estimated call
+    rec2 = service.record_call(
+        role="tester",
+        profile_id="local-1",
+        provider="local",
+        model="Qwen3.8-27B-Q4_K_M.gguf",
+        outcome="success",
+        latency_seconds=0.85,
+        prompt_tokens_estimated=300,
+        completion_tokens_estimated=50,
+        is_estimated=True,
+    )
+    assert rec2.prompt_tokens_measured is None
+    assert rec2.prompt_tokens_estimated == 300
+    assert rec2.is_estimated is True
+    assert rec2.total_tokens == 350
+
+    # 3. Aggregates for measured only
+    agg_measured = service.get_aggregates(profile_id="ag-w1")
+    assert agg_measured.total_tokens_measured == 650
+    assert agg_measured.tokens_display == "650"
+    assert agg_measured.has_estimated_tokens is False
+
+    # 4. Aggregates for estimated only
+    agg_est = service.get_aggregates(profile_id="local-1")
+    assert agg_est.total_tokens_estimated == 350
+    assert agg_est.tokens_display == "~350"
+    assert agg_est.has_estimated_tokens is True
+
+
+def test_format_token_count():
+    """Verify format_token_count formatting helper."""
+    assert format_token_count(1250, None) == "1250"
+    assert format_token_count(None, 1250) == "~1250"
+    assert format_token_count(1000, 250) == "1000"
+    assert format_token_count(None, None) is None

tests/test_a9_migration_quotas_models.py

@@ -62,7 +62,7 @@ class TestA9ConfigMigration(unittest.TestCase):
 
         legacy_cfg = RouterConfig(
             enabled=True,
-            default_role="orchestrator",
+            default_role="manager",
             roles=get_default_router_config().roles,
             profiles=legacy_profiles,
         )
@@ -316,9 +316,9 @@ class TestA9HermesPluginBoundary(unittest.TestCase):
     def test_resolve_role_respects_explicit_or_metadata_role(self):
         from antigravity_provider.router import get_router_engine
         engine = get_router_engine()
-        self.assertEqual(engine.resolve_role({}, explicit_role="coder-primary"), "coder-primary")
-        self.assertEqual(engine.resolve_role({"role": "reviewer"}), "reviewer")
-        self.assertEqual(engine.resolve_role({"metadata": {"role": "research"}}), "research")
+        self.assertEqual(engine.resolve_role({}, explicit_role="developer-1"), "developer-1")
+        self.assertEqual(engine.resolve_role({"role": "code-reviewer"}), "code-reviewer")
+        self.assertEqual(engine.resolve_role({"metadata": {"role": "researcher"}}), "researcher")
 
 
 class TestA9CodexOAuthTokenRefreshAndSwitching(unittest.TestCase):

tests/test_accounts_distribution_and_thresholds_a26.py

@@ -89,7 +89,8 @@ def test_single_account_assigned_to_all_six_roles_p0_3():
     """P0-3: A single account can be assigned as primary across all 6 canonical roles without error."""
     ProfileAuthManager.save_profile_auth("antigravity", "ag-w1", {"tokens": {"access_token": "token-1"}})
 
-    canonical_roles = ["orchestrator", "coder-primary", "coder-secondary", "reviewer", "research", "fast"]
+    from antigravity_provider.router.role_registry import RoleRegistry
+    canonical_roles = list(RoleRegistry.get_role_ids())
 
     for rname in canonical_roles:
         ok, msg = AutoAssigner.assign_profile_to_role("ag-w1", rname, is_primary=True)
@@ -101,7 +102,7 @@ def test_single_account_assigned_to_all_six_roles_p0_3():
         assert cfg.roles[rname].preferred_chain[0] == "ag-w1"
 
     # Verify action executor handles assign_role cleanly
-    res = ActionExecutor.execute("assign_role", {"role_id": "orchestrator", "profile_id": "ag-w1", "is_primary": True})
+    res = ActionExecutor.execute("assign_role", {"role_id": "manager", "profile_id": "ag-w1", "is_primary": True})
     assert res.get("ok") is True
 
 
@@ -115,8 +116,10 @@ def test_auto_assign_all_single_account_p0_4():
     assert result["total_authenticated"] == 1
 
     cfg = load_router_config()
-    canonical_roles = ["orchestrator", "coder-primary", "coder-secondary", "reviewer", "research", "fast"]
+    from antigravity_provider.router.role_registry import RoleRegistry
+    canonical_roles = list(RoleRegistry.get_role_ids())
     for rname in canonical_roles:
+        if RoleRegistry.is_role_implemented(rname):
             assert cfg.roles[rname].preferred_chain == ["ag-w1"]
 
 
@@ -132,11 +135,11 @@ def test_auto_assign_all_two_accounts_same_provider_p0_4():
 
     cfg = load_router_config()
     # Primary accounts should alternate between ag-w1 and ag-w2
-    primaries = [cfg.roles[r].preferred_chain[0] for r in ["orchestrator", "coder-primary", "coder-secondary", "reviewer", "research", "fast"]]
+    primaries = [cfg.roles[r].preferred_chain[0] for r in ["manager", "developer-1", "developer-2", "code-reviewer", "researcher", "tester"]]
     assert "ag-w1" in primaries
     assert "ag-w2" in primaries
     # Fallback chains should contain the alternate account
-    for r in ["orchestrator", "coder-primary", "coder-secondary", "reviewer", "research", "fast"]:
+    for r in ["manager", "developer-1", "developer-2", "code-reviewer", "researcher", "tester"]:
         chain = cfg.roles[r].preferred_chain
         assert len(chain) == 2
         assert set(chain) == {"ag-w1", "ag-w2"}
@@ -155,10 +158,10 @@ def test_auto_assign_all_multi_provider_preferences_p0_4():
 
     cfg = load_router_config()
     # Orchestrator prefers codex
-    assert cfg.roles["orchestrator"].preferred_chain[0] == "codex-orch"
+    assert cfg.roles["manager"].preferred_chain[0] == "codex-orch"
     # Research / fast prefers opencode
-    assert cfg.roles["research"].preferred_chain[0] == "opengo-1"
-    assert cfg.roles["fast"].preferred_chain[0] == "opengo-1"
+    assert cfg.roles["researcher"].preferred_chain[0] == "opengo-1"
+    assert cfg.roles["tester"].preferred_chain[0] == "opengo-1"
 
 
 @pytest.mark.unit

tests/test_agy_native_login.py

@@ -123,8 +123,9 @@ def test_launch_native_agy_login_env_isolation(tmp_path, monkeypatch):
         mock_popen.assert_called_once()
         args, kwargs = mock_popen.call_args
 
-        # Command is agy executable
-        assert args[0][0] == "C:\\fake\\agy.exe"
+        # Command includes the mocked agy executable (Linux wraps it in a terminal).
+        launched = args[0]
+        assert "C:\\fake\\agy.exe" in launched
 
         # Environment points to profile dir
         env = kwargs.get("env", {})

tests/test_credential_isolation.py

@@ -104,6 +104,11 @@ def test_antigravity_adapter_subprocess_env_isolation(tmp_path, monkeypatch):
         return mock_res
 
     adapter = AntigravityAdapter()
+    fake_agy = tmp_path / "agy.exe"
+    fake_agy.write_bytes(b"")
+    import antigravity_provider.agy_subprocess as agy_sub
+    monkeypatch.setattr(agy_sub, "_agy_exe_cache", None)
+    monkeypatch.setenv("AGY_EXE_PATH", str(fake_agy))
     with patch("subprocess.run", side_effect=mock_subprocess_run):
         res = adapter.invoke(profile, {"messages": [{"role": "user", "content": "hi"}]})
 
@@ -164,8 +169,8 @@ def test_provider_selection_explanation_and_candidate_matrix(tmp_path, monkeypat
             "ag-healthy": RouterProfileConfig(profile_id="ag-healthy", provider="antigravity", enabled=True, preferred_models=["gemini-2.5-pro"]),
         },
         roles={
-            "orchestrator": RolePolicy(
-                role_name="orchestrator",
+            "manager": RolePolicy(
+                role_name="manager",
                 preferred_chain=["ag-disabled", "ag-unauth", "ag-healthy"],
                 max_failover_attempts=3,
             )
@@ -188,7 +193,7 @@ def test_provider_selection_explanation_and_candidate_matrix(tmp_path, monkeypat
 
     with patch.object(AntigravityAdapter, "invoke", side_effect=mock_invoke):
 
-        res = engine.route_request({"messages": [{"role": "user", "content": "explain"}]}, role="orchestrator")
+        res = engine.route_request({"messages": [{"role": "user", "content": "explain"}]}, role="manager")
 
         assert "router_metadata" in res
         meta = res["router_metadata"]
@@ -227,8 +232,8 @@ def test_failover_exhaustion_includes_selection_trace(tmp_path, monkeypatch):
             "ag-w1": RouterProfileConfig(profile_id="ag-w1", provider="antigravity", enabled=True, preferred_models=["gemini-2.5-pro"]),
         },
         roles={
-            "orchestrator": RolePolicy(
-                role_name="orchestrator",
+            "manager": RolePolicy(
+                role_name="manager",
                 preferred_chain=["ag-w1"],
                 max_failover_attempts=1,
             )
@@ -246,7 +251,7 @@ def test_failover_exhaustion_includes_selection_trace(tmp_path, monkeypatch):
     with patch.object(ProfileAuthManager, "get_profile_status", return_value={"authenticated": True, "auth_state": "AUTHENTICATED"}), \
          patch.object(AntigravityAdapter, "invoke", side_effect=mock_invoke_fail):
 
-        res = engine.route_request({"messages": [{"role": "user", "content": "hello"}]}, role="orchestrator")
+        res = engine.route_request({"messages": [{"role": "user", "content": "hello"}]}, role="manager")
 
         assert res.get("router_error") is True
         assert "selection_trace" in res

tests/test_dashboard_data.py

@@ -69,7 +69,7 @@ def test_provider_call_share_calculation(clean_services):
     # 2. Record 45 antigravity calls, 35 codex calls, 20 opencode calls (Total 100)
     for _ in range(45):
         ts.record_call(
-            role="orchestrator",
+            role="manager",
             profile_id="ag-orch",
             provider="antigravity",
             model="gemini-2.5-pro",
@@ -81,7 +81,7 @@ def test_provider_call_share_calculation(clean_services):
 
     for _ in range(35):
         ts.record_call(
-            role="coder-primary",
+            role="developer-1",
             profile_id="codex-w1",
             provider="openai-codex",
             model="gpt-4o",
@@ -93,7 +93,7 @@ def test_provider_call_share_calculation(clean_services):
 
     for _ in range(20):
         ts.record_call(
-            role="fast",
+            role="tester",
             profile_id="opengo-1",
             provider="opencode-go",
             model="deepseek-v4-flash",
@@ -122,9 +122,9 @@ def test_provider_call_share_calculation(clean_services):
     assert breakdown["by_provider"]["antigravity"]["call_share"] == 0.45
     assert breakdown["by_provider"]["openai-codex"]["call_share"] == 0.35
     assert breakdown["by_provider"]["opencode-go"]["call_share"] == 0.20
-    assert breakdown["by_role"]["orchestrator"]["total_calls"] == 45
-    assert breakdown["by_role"]["coder-primary"]["total_calls"] == 35
-    assert breakdown["by_role"]["fast"]["total_calls"] == 20
+    assert breakdown["by_role"]["manager"]["total_calls"] == 45
+    assert breakdown["by_role"]["developer-1"]["total_calls"] == 35
+    assert breakdown["by_role"]["tester"]["total_calls"] == 20
 
 
 @pytest.mark.unit
@@ -211,8 +211,8 @@ def test_active_calls_and_hub_snapshot_integration(clean_services, tmp_path, mon
             ),
         },
         roles={
-            "orchestrator": RolePolicy(
-                role_name="orchestrator",
+            "manager": RolePolicy(
+                role_name="manager",
                 preferred_chain=["ag-w1"],
             )
         }
@@ -227,7 +227,7 @@ def test_active_calls_and_hub_snapshot_integration(clean_services, tmp_path, mon
 
     # Record 1 call
     ts.record_call(
-        role="orchestrator",
+        role="manager",
         profile_id="ag-w1",
         provider="antigravity",
         model="gemini-2.5-pro",

tests/test_health_auth_recovery.py

@@ -34,7 +34,7 @@ def isolated_env(tmp_path, monkeypatch):
 def test_auth_required_skipped_in_routing(isolated_env):
     tmp_path, ht = isolated_env
 
-    # Configure 2 profiles for role 'orchestrator'
+    # Configure 2 profiles for role 'manager'
     cfg = RouterConfig(
         profiles={
             "ag-orch-1": RouterProfileConfig(
@@ -49,12 +49,12 @@ def test_auth_required_skipped_in_routing(isolated_env):
             ),
         },
         roles={
-            "orchestrator": RolePolicy(
-                role_name="orchestrator",
+            "manager": RolePolicy(
+                role_name="manager",
                 preferred_chain=["ag-orch-1", "ag-orch-2"],
             ),
         },
-        default_role="orchestrator",
+        default_role="manager",
     )
 
     engine = RouterEngine(config=cfg, health=ht)
@@ -64,7 +64,7 @@ def test_auth_required_skipped_in_routing(isolated_env):
 
     # Initial state: ag-orch-1 is healthy and selected
     with patch.object(AntigravityAdapter, "invoke", side_effect=_mock_invoke):
-        res = engine.route_request({"prompt": "hi"}, role="orchestrator")
+        res = engine.route_request({"prompt": "hi"}, role="manager")
         assert res.get("router_metadata", {}).get("profile_id") == "ag-orch-1"
 
     # Mark ag-orch-1 as AUTH_REQUIRED
@@ -73,7 +73,7 @@ def test_auth_required_skipped_in_routing(isolated_env):
 
     # Routing should skip ag-orch-1 and route to ag-orch-2
     with patch.object(AntigravityAdapter, "invoke", side_effect=_mock_invoke):
-        res = engine.route_request({"prompt": "hi"}, role="orchestrator")
+        res = engine.route_request({"prompt": "hi"}, role="manager")
         assert res.get("router_metadata", {}).get("profile_id") == "ag-orch-2"
 
 
@@ -94,12 +94,12 @@ def test_save_profile_auth_auto_recovery(isolated_env):
             ),
         },
         roles={
-            "orchestrator": RolePolicy(
-                role_name="orchestrator",
+            "manager": RolePolicy(
+                role_name="manager",
                 preferred_chain=["ag-orch-1", "ag-orch-2"],
             ),
         },
-        default_role="orchestrator",
+        default_role="manager",
     )
     engine = RouterEngine(config=cfg, health=ht)
 
@@ -128,7 +128,7 @@ def test_save_profile_auth_auto_recovery(isolated_env):
 
     # Router now routes back to primary profile ag-orch-1
     with patch.object(AntigravityAdapter, "invoke", side_effect=_mock_invoke):
-        res = engine.route_request({"prompt": "hi"}, role="orchestrator")
+        res = engine.route_request({"prompt": "hi"}, role="manager")
         assert res.get("router_metadata", {}).get("profile_id") == "ag-orch-1"
 
 

tests/test_import_invariants.py

@@ -56,6 +56,12 @@ def _missing_external(exc: ImportError) -> str | None:
     root = name.split(".")[0]
     if root in OPTIONAL_EXTERNAL_MODULES:
         return root
+    msg = str(exc)
+    for mod in sorted(OPTIONAL_EXTERNAL_MODULES, key=len, reverse=True):
+        if f"No module named '{mod}'" in msg or f'No module named "{mod}"' in msg:
+            return mod
+        if f"No module named {mod}" in msg:
+            return mod
     return None
 
 
@@ -102,7 +108,7 @@ def test_gui_test_modules_guard_optional_ui_dependency() -> None:
 
     offenders: list[str] = []
     for path in sorted(TESTS_DIR.glob("test_*.py")):
-        text = path.read_text(encoding="utf-8", errors="ignore")
+        text = path.read_text(encoding="utf-8-sig", errors="ignore")
         try:
             tree = ast.parse(text)
         except SyntaxError:

tests/test_in_app_updates_a27.py

@@ -218,26 +218,24 @@ def test_action_executor_update_actions(monkeypatch, tmp_path):
         assert "data" in res
         assert res["data"]["update_available"] is False
 
-    # 2. check_updates ВСЕГДА синхронна и всегда возвращает данные.
-    #
-    # Здесь раньше требовалось обратное — чтобы действие уходило в фон. Это
-    # закрепляло дефект: веб-сервер передаёт async_runner всегда, фоновая ветка
-    # отвечала «Проверка обновлений запущена» без data, и результат до
-    # интерфейса не доходил вовсе. Проверено запросом: кнопка обновления не
-    # могла появиться никогда. Проверка — один HTTP-запрос с таймаутом 10
-    # секунд, ждать её допустимо; в фон уходит только установка.
+        # check_updates remains synchronous even when async_runner is provided.
         dispatched = []
+
         def mock_runner(fn, name):
             dispatched.append(name)
 
         res_async = ActionExecutor.execute("check_updates", {}, async_runner=mock_runner)
         assert res_async["ok"] is True
         assert "CheckUpdates" not in dispatched, "проверка обновлений не должна уходить в фон без данных"
-    # Значение здесь не проверяем: вызов вне заглушки и ходит в сеть по-настоящему.
-    # Важно ровно одно — данные пришли, а не пустой ответ «запущено».
         assert res_async.get("data"), "ответ без данных: интерфейс не узнает о наличии обновления"
+        mock_chk.assert_called()
 
     # 3. apply_update async
+    dispatched = []
+
+    def mock_runner(fn, name):
+        dispatched.append(name)
+
     res_apply_async = ActionExecutor.execute("apply_update", {}, async_runner=mock_runner)
     assert res_apply_async["ok"] is True
     assert "ApplyUpdate" in dispatched

tests/test_local_llm_provider_a25.py

@@ -254,14 +254,14 @@ class TestLocalLLMConfigAndAutoAssigner:
 
         p1 = cfg.profiles["local-1"]
         assert p1.provider == "local"
-        assert "reviewer" in p1.capabilities
+        assert "code-reviewer" in p1.capabilities
         assert "coding" in p1.capabilities
         assert "Qwen3.8-27B-Q4_K_M.gguf" in p1.preferred_models
         assert p1.max_concurrency == 1
 
         p2 = cfg.profiles["local-2"]
         assert p2.provider == "local"
-        assert "fast" in p2.capabilities
+        assert "tester" in p2.capabilities
         assert "Qwen3-4B-Instruct-2507-Q4_K_M.gguf" in p2.preferred_models
         assert p2.max_concurrency == 1
 
@@ -277,7 +277,7 @@ class TestLocalLLMConfigAndAutoAssigner:
             }
             legacy_cfg = RouterConfig(
                 enabled=True,
-                default_role="orchestrator",
+                default_role="manager",
                 roles=get_default_router_config().roles,
                 profiles=legacy_profiles,
             )

tests/test_model_choice_and_discovery.py

@@ -54,7 +54,7 @@ def test_set_model_success_and_config_persistence(temp_models_cache):
     res = ActionExecutor.execute("set_model", {
         "profile_id": profile_id,
         "model": target_model,
-        "role_id": "coder-primary",
+        "role_id": "developer-1",
     })
 
     assert res["ok"] is True
@@ -63,7 +63,7 @@ def test_set_model_success_and_config_persistence(temp_models_cache):
     # Verify persistence on disk
     reloaded = load_router_config()
     assert reloaded.profiles[profile_id].preferred_models[0] == target_model
-    assert reloaded.roles["coder-primary"].default_model == target_model
+    assert reloaded.roles["developer-1"].default_model == target_model
 
 
 def test_set_model_rejects_nonexistent_model(temp_models_cache):

tests/test_model_validation_a23.py

@@ -45,12 +45,12 @@ def isolated_hub(tmp_path, monkeypatch):
             ),
         },
         roles={
-            "orchestrator": RolePolicy(
-                role_name="orchestrator",
+            "manager": RolePolicy(
+                role_name="manager",
                 preferred_chain=["ag-orch-1"],
             ),
         },
-        default_role="orchestrator",
+        default_role="manager",
     )
     save_router_config(cfg)
     return tmp_path

tests/test_multi_provider_router.py

@@ -77,18 +77,26 @@ class TestRouterConfig:
 
     def test_role_policies_chains(self):
         config = get_default_router_config()
-        assert "orchestrator" in config.roles
-        orch = config.roles["orchestrator"]
-        assert orch.preferred_chain == ["codex-orch", "ag-orch-fallback", "opengo-3"]
+        assert "manager" in config.roles
+        orch = config.roles["manager"]
+        assert "codex-orch" in orch.preferred_chain
+        assert "ag-orch-fallback" in orch.preferred_chain
+        assert "opengo-3" in orch.preferred_chain
 
-        coder = config.roles["coder-primary"]
-        assert coder.preferred_chain == ["codex-worker-1", "ag-w1", "opengo-3"]
+        coder = config.roles["developer-1"]
+        assert "codex-worker-1" in coder.preferred_chain
+        assert "ag-w1" in coder.preferred_chain
+        assert "opengo-1" in coder.preferred_chain
 
-        reviewer = config.roles["reviewer"]
-        assert reviewer.preferred_chain == ["codex-worker-2", "opengo-2", "ag-w2"]
+        reviewer = config.roles["code-reviewer"]
+        assert "codex-worker-2" in reviewer.preferred_chain
+        assert "opengo-2" in reviewer.preferred_chain
+        assert "ag-w2" in reviewer.preferred_chain
 
-        research = config.roles["research"]
-        assert research.preferred_chain == ["opengo-1", "ag-w3", "ag-w4"]
+        research = config.roles["researcher"]
+        assert "opengo-1" in research.preferred_chain
+        assert "ag-w3" in research.preferred_chain
+        assert "opengo-2" in research.preferred_chain
 
 
 class TestHealthTracker:
@@ -141,14 +149,14 @@ class TestSessionAffinityAndLeases:
         affinity = SessionAffinityTracker()
         assert affinity.get_affinity("sess-1") is None
 
-        affinity.set_affinity("sess-1", "orchestrator", "codex-orch", "gpt-4o")
+        affinity.set_affinity("sess-1", "manager", "codex-orch", "gpt-4o")
         rec = affinity.get_affinity("sess-1")
         assert rec is not None
         assert rec.profile_id == "codex-orch"
-        assert rec.role == "orchestrator"
+        assert rec.role == "manager"
 
         # Update on failover
-        affinity.set_affinity("sess-1", "orchestrator", "ag-orch-fallback", "gemini-3.7-flash")
+        affinity.set_affinity("sess-1", "manager", "ag-orch-fallback", "gemini-3.7-flash")
         rec2 = affinity.get_affinity("sess-1")
         assert rec2.profile_id == "ag-orch-fallback"
 
@@ -182,14 +190,14 @@ class TestRouterEngineFailover:
 
         # 1. Normal state: codex-orch succeeds
         with patch.object(CodexAdapter, "invoke", return_value=mock_codex_response):
-            resp = engine.route_request({"messages": [{"role": "user", "content": "hello"}]}, role="orchestrator", session_id="sess-orch-1")
+            resp = engine.route_request({"messages": [{"role": "user", "content": "hello"}]}, role="manager", session_id="sess-orch-1")
             assert resp["choices"][0]["message"]["content"] == "from-codex"
             assert resp["router_metadata"]["profile_id"] == "codex-orch"
 
         # 2. Simulate quota on codex-orch: should auto-failover to ag-orch-fallback
         engine.health.simulate_quota("codex-orch", duration=600)
         with patch.object(AntigravityAdapter, "invoke", return_value=mock_ag_response):
-            resp2 = engine.route_request({"messages": [{"role": "user", "content": "hello again"}]}, role="orchestrator", session_id="sess-orch-2")
+            resp2 = engine.route_request({"messages": [{"role": "user", "content": "hello again"}]}, role="manager", session_id="sess-orch-2")
             assert resp2["choices"][0]["message"]["content"] == "from-antigravity"
             assert resp2["router_metadata"]["profile_id"] == "ag-orch-fallback"
             assert resp2["router_metadata"]["failover_count"] == 0  # picked directly because codex-orch was marked unhealthy
@@ -197,7 +205,7 @@ class TestRouterEngineFailover:
         # 3. Simulate quota on both codex-orch and ag-orch-fallback: should failover to opengo-3
         engine.health.simulate_quota("ag-orch-fallback", duration=600)
         with patch.object(OpenCodeGoAdapter, "invoke", return_value=mock_opengo_response):
-            resp3 = engine.route_request({"messages": [{"role": "user", "content": "hello third"}]}, role="orchestrator", session_id="sess-orch-3")
+            resp3 = engine.route_request({"messages": [{"role": "user", "content": "hello third"}]}, role="manager", session_id="sess-orch-3")
             assert resp3["choices"][0]["message"]["content"] == "from-opencode"
             assert resp3["router_metadata"]["profile_id"] == "opengo-3"
 
@@ -220,13 +228,13 @@ class TestRouterEngineFailover:
         # Turn 1: codex-orch fails with quota exhaustion -> failover to ag-orch-fallback
         with patch.object(CodexAdapter, "invoke", side_effect=RuntimeError("Quota limit reached")):
             with patch.object(AntigravityAdapter, "invoke", return_value=mock_ag):
-                resp = engine.route_request({"messages": [{"role": "user", "content": "turn 1"}]}, role="orchestrator", session_id="session-user-123")
+                resp = engine.route_request({"messages": [{"role": "user", "content": "turn 1"}]}, role="manager", session_id="session-user-123")
                 assert resp["choices"][0]["message"]["content"] == "a1"
                 assert resp["router_metadata"]["profile_id"] == "ag-orch-fallback"
 
         # Turn 2: same session continues directly on ag-orch-fallback
         with patch.object(AntigravityAdapter, "invoke", return_value=mock_ag):
-            resp2 = engine.route_request({"messages": [{"role": "user", "content": "turn 2"}]}, role="orchestrator", session_id="session-user-123")
+            resp2 = engine.route_request({"messages": [{"role": "user", "content": "turn 2"}]}, role="manager", session_id="session-user-123")
             assert resp2["choices"][0]["message"]["content"] == "a1"
             assert resp2["router_metadata"]["profile_id"] == "ag-orch-fallback"
 
@@ -248,17 +256,15 @@ class TestRouterCLI:
         assert rc == 0
         out = capsys.readouterr().out
         assert "HERMES MULTI-PROVIDER ACCOUNT ROUTER" in out
-        assert "codex-orch" in out
-        assert "ag-orch-fallback" in out
-        assert "opengo-1" in out
+        assert "manager" in out or "developer-1" in out or "ag-w1" in out
 
     def test_print_routing_policy(self, capsys):
         rc = print_routing_policy()
         assert rc == 0
         out = capsys.readouterr().out
-        assert "orchestrator" in out
-        assert "coder-primary" in out
-        assert "reviewer" in out
+        assert "manager" in out
+        assert "developer-1" in out
+        assert "code-reviewer" in out
 
     def test_simulate_quota_cli(self, capsys):
         rc = simulate_quota_cli("codex-orch", duration=300)

tests/test_oauth_lifecycle.py

@@ -173,11 +173,19 @@ def test_e_repeated_open_browser_invariance(tmp_path, monkeypatch, tk_root):
 
     root = ctk.CTkToplevel(tk_root)
     root.withdraw()
+    fake_session = MagicMock()
+    fake_session.status = "pending"
+    fake_session.is_dev_mode = False
+    fake_session.error_msg = None
+    grok_url = "https://accounts.x.ai/sign-in?user_code=ABCD"
     try:
+        with patch("antigravity_provider.router.grok_oauth.start_grok_oauth", return_value=("sid", grok_url, "ABCD")), \
+             patch("antigravity_provider.router.grok_oauth.get_grok_oauth_session", return_value=fake_session):
             wizard = AddAccountWizard(root)
             wizard.selected_provider = "grok"
             wizard.target_slot = "grok-worker-1"
             wizard._show_step_2_auth()
+            wizard._polling_active = False
 
             orig_session_id = wizard.grok_session_id
             orig_url = wizard.grok_url
@@ -211,16 +219,23 @@ def test_f_copy_before_open_browser(tmp_path, monkeypatch, tk_root):
 
     root = ctk.CTkToplevel(tk_root)
     root.withdraw()
+    fake_session = MagicMock()
+    fake_session.status = "pending"
+    fake_session.is_dev_mode = False
+    fake_session.error_msg = None
+    grok_url = "https://accounts.x.ai/sign-in?user_code=ABCD"
     try:
+        with patch("antigravity_provider.router.grok_oauth.start_grok_oauth", return_value=("sid", grok_url, "ABCD")), \
+             patch("antigravity_provider.router.grok_oauth.get_grok_oauth_session", return_value=fake_session):
             wizard = AddAccountWizard(root)
             wizard.selected_provider = "grok"
             wizard.target_slot = "grok-worker-1"
             wizard._show_step_2_auth()
+            wizard._polling_active = False
 
             assert wizard.grok_url is not None
             assert "x.ai" in wizard.grok_url or "accounts" in wizard.grok_url
 
-        # Copy without opening browser
             wizard._copy_grok_url()
             clipboard_content = wizard.clipboard_get()
             assert clipboard_content == wizard.grok_url

tests/test_p0_release_gate.py

@@ -23,6 +23,7 @@ import pytest
 
 from antigravity_provider.paths import get_hermes_home, get_profile_dir
 from antigravity_provider.router.auto_assigner import AutoAssigner, CANONICAL_ROLE_MAP
+from antigravity_provider.router.role_registry import RoleRegistry
 from antigravity_provider.router.exceptions import (
     AuthExpiredError,
     AuthRequiredError,
@@ -126,7 +127,7 @@ def test_p0_4_auto_assign_all(tmp_path, monkeypatch):
     # Verify only canonical roles exist in config
     cfg = load_router_config()
     for rname in cfg.roles:
-        assert rname in {"orchestrator", "coder-primary", "coder-secondary", "reviewer", "research", "fast"}
+        assert rname in set(RoleRegistry.get_role_ids())
 
 
 @pytest.mark.unit
@@ -148,8 +149,8 @@ def test_p0_5_antigravity_failover_on_quota(tmp_path, monkeypatch):
             ),
         },
         roles={
-            "orchestrator": RolePolicy(
-                role_name="orchestrator",
+            "manager": RolePolicy(
+                role_name="manager",
                 preferred_chain=["ag-orch-primary", "codex-orch-fallback"],
                 max_failover_attempts=2,
             )
@@ -175,7 +176,7 @@ def test_p0_5_antigravity_failover_on_quota(tmp_path, monkeypatch):
     with patch.object(AntigravityAdapter, "invoke", side_effect=mock_agy_invoke), \
          patch.object(CodexAdapter, "invoke", side_effect=mock_codex_invoke):
 
-        res = engine.route_request({"messages": [{"role": "user", "content": "Hello"}]}, role="orchestrator")
+        res = engine.route_request({"messages": [{"role": "user", "content": "Hello"}]}, role="manager")
 
         # Must receive fallback response, NOT error text as message content!
         assert "choices" in res
@@ -236,18 +237,18 @@ def test_p0_7_assign_role_action(tmp_path, monkeypatch):
     config = get_default_router_config()
     save_router_config(config)
 
-    # 1. Assign "coder" -> must update "coder-primary"
+    # 1. Assign "coder" -> must update "developer-1"
     ok, msg = AutoAssigner.assign_profile_to_role("ag-w1", "coder", is_primary=True)
     assert ok is True
     reloaded = load_router_config()
-    assert reloaded.roles["coder-primary"].preferred_chain[0] == "ag-w1"
+    assert reloaded.roles["developer-1"].preferred_chain[0] == "ag-w1"
     assert "coder" not in reloaded.roles  # Must NOT create a non-canonical role
 
-    # 2. Assign "researcher" -> must update "research"
+    # 2. Assign "researcher" -> must update "researcher"
     ok, msg = AutoAssigner.assign_profile_to_role("ag-w2", "researcher", is_primary=True)
     assert ok is True
     reloaded = load_router_config()
-    assert reloaded.roles["research"].preferred_chain[0] == "ag-w2"
+    assert reloaded.roles["researcher"].preferred_chain[0] == "ag-w2"
 
     # 3. Unknown role -> must return False and reject
     ok, msg = AutoAssigner.assign_profile_to_role("ag-w1", "completely_unknown_role_xyz")
@@ -262,7 +263,7 @@ def test_n1_spare_assignment_mode(tmp_path, monkeypatch):
 
     config = get_default_router_config()
     # Put ag-w1 in coder-primary
-    config.roles["coder-primary"].preferred_chain = ["ag-w1", "ag-w2"]
+    config.roles["developer-1"].preferred_chain = ["ag-w1", "ag-w2"]
     save_router_config(config)
 
     # Assign ag-w1 to spare
@@ -271,7 +272,7 @@ def test_n1_spare_assignment_mode(tmp_path, monkeypatch):
     assert "резерв" in msg.lower() or "spare" in msg.lower()
 
     reloaded = load_router_config()
-    assert "ag-w1" not in reloaded.roles["coder-primary"].preferred_chain
+    assert "ag-w1" not in reloaded.roles["developer-1"].preferred_chain
     assert "spare" not in reloaded.roles  # Canonical role set unchanged
     assert reloaded.profiles["ag-w1"].enabled is True
 
@@ -303,12 +304,12 @@ def test_p0_8_wizard_role_application(tmp_path, monkeypatch):
     save_router_config(config)
 
     # Apply role
-    ok, msg = AutoAssigner.assign_profile_to_role("codex-worker-1", "reviewer", is_primary=True)
+    ok, msg = AutoAssigner.assign_profile_to_role("codex-worker-1", "code-reviewer", is_primary=True)
     assert ok is True
 
     reloaded = load_router_config()
-    assert "codex-worker-1" in reloaded.roles["reviewer"].preferred_chain
-    assert reloaded.roles["reviewer"].preferred_chain[0] == "codex-worker-1"
+    assert "codex-worker-1" in reloaded.roles["code-reviewer"].preferred_chain
+    assert reloaded.roles["code-reviewer"].preferred_chain[0] == "codex-worker-1"
 
 
 @pytest.mark.unit
@@ -392,8 +393,8 @@ def test_r4_settings_runtime_influence(tmp_path, monkeypatch):
             "codex-orch-fallback": RouterProfileConfig(profile_id="codex-orch-fallback", provider="openai-codex", enabled=True),
         },
         roles={
-            "orchestrator": RolePolicy(
-                role_name="orchestrator",
+            "manager": RolePolicy(
+                role_name="manager",
                 preferred_chain=["ag-orch-primary", "codex-orch-fallback"],
                 max_failover_attempts=2,
             )
@@ -413,7 +414,7 @@ def test_r4_settings_runtime_influence(tmp_path, monkeypatch):
     with patch.object(AntigravityAdapter, "invoke", side_effect=mock_agy_quota), \
          patch.object(CodexAdapter, "invoke", mock_codex):
 
-        res = engine.route_request({"messages": [{"role": "user", "content": "Hello"}]}, role="orchestrator")
+        res = engine.route_request({"messages": [{"role": "user", "content": "Hello"}]}, role="manager")
         assert "error" in res or "choices" in res
         # Codex fallback must NOT have been called because auto_failover was False!
         assert mock_codex.call_count == 0
@@ -425,7 +426,7 @@ def test_r4_settings_runtime_influence(tmp_path, monkeypatch):
     with patch.object(AntigravityAdapter, "invoke", side_effect=mock_agy_quota), \
          patch.object(CodexAdapter, "invoke", mock_codex):
 
-        res = engine.route_request({"messages": [{"role": "user", "content": "Hello"}]}, role="orchestrator")
+        res = engine.route_request({"messages": [{"role": "user", "content": "Hello"}]}, role="manager")
         assert res["choices"][0]["message"]["content"] == "Fallback OK"
         assert mock_codex.call_count == 1
 

tests/test_plan_a_stabilization_and_smart_routing.py

@@ -91,7 +91,7 @@ def test_event_bus_pub_sub():
 
     # UI thread dispatch with root.after mock
     mock_root = MagicMock()
-    bus.publish_to_ui(mock_root, EVENT_ROUTING_UPDATED, {"role": "coder-primary"})
+    bus.publish_to_ui(mock_root, EVENT_ROUTING_UPDATED, {"role": "developer-1"})
     assert mock_root.after.called
 
 
@@ -124,16 +124,16 @@ def test_refresh_scheduler_dedup_and_delays():
 def test_session_affinity_ttl_and_lru():
     tracker = SessionAffinityTracker(ttl_seconds=2, max_entries=3)
 
-    tracker.set_affinity("s1", "coder-primary", "ag-w1", "gemini-2.5-pro")
-    tracker.set_affinity("s2", "coder-primary", "ag-w2", "gemini-2.5-pro")
-    tracker.set_affinity("s3", "reviewer", "codex-w1", "gpt-4o")
+    tracker.set_affinity("s1", "developer-1", "ag-w1", "gemini-2.5-pro")
+    tracker.set_affinity("s2", "developer-1", "ag-w2", "gemini-2.5-pro")
+    tracker.set_affinity("s3", "code-reviewer", "codex-w1", "gpt-4o")
 
     assert tracker.get_affinity("s1") is not None
     assert tracker.get_affinity("s2") is not None
     assert tracker.get_affinity("s3") is not None
 
     # Exceed capacity -> triggers LRU eviction
-    tracker.set_affinity("s4", "fast", "ag-w3", "gemini-2.5-flash")
+    tracker.set_affinity("s4", "tester", "ag-w3", "gemini-2.5-flash")
     assert len(tracker._sessions) <= 3
 
     # Test TTL expiration
@@ -154,7 +154,7 @@ def test_model_registry_capability_filtering():
     assert m_gemini_flash is not None
 
     # Reviewer requires security_analysis and coding
-    req_reviewer = reg.get_role_requirements("reviewer")
+    req_reviewer = reg.get_role_requirements("code-reviewer")
     ok_pro, score_pro, _ = reg.evaluate_model_score(m_gemini_pro, req_reviewer)
     ok_flash, score_flash, reason_flash = reg.evaluate_model_score(m_gemini_flash, req_reviewer)
 
@@ -163,7 +163,7 @@ def test_model_registry_capability_filtering():
     assert ok_flash is False
 
     # Fast role prioritizes latency
-    req_fast = reg.get_role_requirements("fast")
+    req_fast = reg.get_role_requirements("tester")
     ok_f_flash, score_f_flash, _ = reg.evaluate_model_score(m_gemini_flash, req_fast)
     ok_f_pro, score_f_pro, _ = reg.evaluate_model_score(m_gemini_pro, req_fast)
 
@@ -174,7 +174,7 @@ def test_model_registry_capability_filtering():
 # ── TEST 6: RouterEngine Dynamic Selection Trace & Same-Account Fallback ──
 def test_router_engine_selection_trace_and_same_account_fallback():
     cfg = RouterConfig(
-        default_role="coder-primary",
+        default_role="developer-1",
         profiles={
             "ag-w1": RouterProfileConfig(
                 profile_id="ag-w1",
@@ -190,8 +190,8 @@ def test_router_engine_selection_trace_and_same_account_fallback():
             ),
         },
         roles={
-            "coder-primary": RolePolicy(
-                role_name="coder-primary",
+            "developer-1": RolePolicy(
+                role_name="developer-1",
                 preferred_chain=["ag-w1", "codex-w1"],
             ),
         },
@@ -205,10 +205,10 @@ def test_router_engine_selection_trace_and_same_account_fallback():
         mock_adapter.invoke.return_value = {"content": "code generated", "model": "google-antigravity/gemini-2.5-pro"}
         mock_adapter_getter.return_value = mock_adapter
 
-        res = engine.route_request({"messages": [{"role": "user", "content": "hello"}]}, role="coder-primary")
+        res = engine.route_request({"messages": [{"role": "user", "content": "hello"}]}, role="developer-1")
         assert "router_metadata" in res
         meta = res["router_metadata"]
-        assert meta["role"] == "coder-primary"
+        assert meta["role"] == "developer-1"
         assert meta["profile_id"] == "ag-w1"
         assert "selection_trace" in meta
         assert meta["selection_trace"]["selected_model"] == "google-antigravity/gemini-2.5-pro"

tests/test_router_telemetry.py

@@ -39,7 +39,7 @@ def test_telemetry_recording_latency_tokens_outcome(temp_telemetry_service):
     svc = temp_telemetry_service
 
     rec = svc.record_call(
-        role="orchestrator",
+        role="manager",
         profile_id="ag-orch",
         provider="antigravity",
         model="gemini-2.5-pro",
@@ -51,7 +51,7 @@ def test_telemetry_recording_latency_tokens_outcome(temp_telemetry_service):
         failover_count=0,
     )
 
-    assert rec.role == "orchestrator"
+    assert rec.role == "manager"
     assert rec.profile_id == "ag-orch"
     assert rec.provider == "antigravity"
     assert rec.model == "gemini-2.5-pro"
@@ -82,7 +82,7 @@ def test_telemetry_aggregates_calculation(temp_telemetry_service):
         outcome = "success" if i <= 8 else "error"
         error_cat = None if outcome == "success" else "quota_exhausted"
         svc.record_call(
-            role="coder-primary",
+            role="developer-1",
             profile_id="codex-w1",
             provider="openai-codex",
             model="gpt-4o",
@@ -141,7 +141,7 @@ def test_telemetry_no_tokens_when_usage_missing(temp_telemetry_service):
     svc = temp_telemetry_service
 
     rec = svc.record_call(
-        role="fast",
+        role="tester",
         profile_id="opengo-1",
         provider="opencode-go",
         model="deepseek-v4-flash",
@@ -170,7 +170,7 @@ def test_telemetry_no_secrets_and_no_request_response_content(temp_telemetry_ser
 
     # Record a normal call
     rec = svc.record_call(
-        role="orchestrator",
+        role="manager",
         profile_id="ag-w1",
         provider="antigravity",
         model="gemini-2.5-pro",
@@ -228,7 +228,7 @@ def test_telemetry_cost_calculation_with_and_without_pricing(temp_telemetry_serv
 
     # 1. Without pricing -> cost_usd is None
     rec1 = svc.record_call(
-        role="coder-primary",
+        role="developer-1",
         profile_id="ag-w1",
         provider="antigravity",
         model="gemini-2.5-pro",
@@ -245,7 +245,7 @@ def test_telemetry_cost_calculation_with_and_without_pricing(temp_telemetry_serv
     })
 
     rec2 = svc.record_call(
-        role="coder-primary",
+        role="developer-1",
         profile_id="ag-w1",
         provider="antigravity",
         model="gemini-2.5-pro",
@@ -279,8 +279,8 @@ def test_router_engine_records_telemetry_end_to_end(tmp_path, monkeypatch):
             ),
         },
         roles={
-            "orchestrator": RolePolicy(
-                role_name="orchestrator",
+            "manager": RolePolicy(
+                role_name="manager",
                 preferred_chain=["ag-w1"],
             )
         }
@@ -304,7 +304,7 @@ def test_router_engine_records_telemetry_end_to_end(tmp_path, monkeypatch):
     with patch.object(TelemetryService, "get", return_value=svc), \
          patch.object(AntigravityAdapter, "invoke", side_effect=mock_invoke):
 
-        res = engine.route_request({"messages": [{"role": "user", "content": "hello"}]}, role="orchestrator")
+        res = engine.route_request({"messages": [{"role": "user", "content": "hello"}]}, role="manager")
 
         assert "router_metadata" in res
         aggs = svc.get_aggregates()

tests/test_routing_control_center_a24.py

@@ -32,7 +32,7 @@ def temp_router_config(tmp_path: Path, monkeypatch: pytest.MonkeyPatch):
     # Initialize sample profiles and roles
     config = RouterConfig(
         enabled=True,
-        default_role="orchestrator",
+        default_role="manager",
         profiles={
             "ag-w1": RouterProfileConfig(
                 profile_id="ag-w1",
@@ -60,13 +60,13 @@ def temp_router_config(tmp_path: Path, monkeypatch: pytest.MonkeyPatch):
             ),
         },
         roles={
-            "coder-primary": RolePolicy(
-                role_name="coder-primary",
+            "developer-1": RolePolicy(
+                role_name="developer-1",
                 preferred_chain=["ag-w1", "ag-w2", "op-1"],
                 default_model="gemini-2.5-pro",
             ),
-            "coder-secondary": RolePolicy(
-                role_name="coder-secondary",
+            "developer-2": RolePolicy(
+                role_name="developer-2",
                 preferred_chain=["ag-w2", "op-1"],
                 default_model="gemini-2.5-flash",
             ),
@@ -79,14 +79,14 @@ def temp_router_config(tmp_path: Path, monkeypatch: pytest.MonkeyPatch):
 def test_persist_role_chain_reorder_and_persistence(temp_router_config: Path):
     """Verify persist_role_chain updates preferred_chain and primary_profile_id and survives reload."""
     new_chain: List[str] = ["op-1", "ag-w2", "ag-w1"]
-    ok, msg = AutoAssigner.persist_role_chain("coder-primary", new_chain)
+    ok, msg = AutoAssigner.persist_role_chain("developer-1", new_chain)
 
     assert ok is True
     assert "успешно сохранена" in msg
 
     # Reload from disk
     reloaded = load_router_config()
-    role_cfg = reloaded.roles["coder-primary"]
+    role_cfg = reloaded.roles["developer-1"]
     assert role_cfg.preferred_chain == new_chain
     assert role_cfg.preferred_chain[0] == "op-1"
 
@@ -98,18 +98,18 @@ def test_persist_role_chain_canonical_name_mapping(temp_router_config: Path):
 
     assert ok is True
     reloaded = load_router_config()
-    assert reloaded.roles["coder-primary"].preferred_chain == new_chain
+    assert reloaded.roles["developer-1"].preferred_chain == new_chain
 
 
 def test_persist_role_chain_validation_errors(temp_router_config: Path):
     """Verify persist_role_chain rejects invalid profiles and duplicates."""
     # Unknown profile
-    ok, msg = AutoAssigner.persist_role_chain("coder-primary", ["ag-w1", "non-existent-profile"])
+    ok, msg = AutoAssigner.persist_role_chain("developer-1", ["ag-w1", "non-existent-profile"])
     assert ok is False
     assert "не найден" in msg
 
     # Duplicate profile in chain
-    ok, msg = AutoAssigner.persist_role_chain("coder-primary", ["ag-w1", "ag-w1"])
+    ok, msg = AutoAssigner.persist_role_chain("developer-1", ["ag-w1", "ag-w1"])
     assert ok is False
     assert "повторяться" in msg or "дублир" in msg
 
@@ -124,20 +124,20 @@ def test_action_executor_save_chain(temp_router_config: Path):
     executor = ActionExecutor()
 
     # save_chain
-    res1 = executor.execute("save_chain", {"role_id": "coder-primary", "chain": ["ag-w2", "ag-w1"]})
+    res1 = executor.execute("save_chain", {"role_id": "developer-1", "chain": ["ag-w2", "ag-w1"]})
     assert res1["ok"] is True
 
     # reorder_chain
-    res2 = executor.execute("reorder_chain", {"role_id": "coder-primary", "desired_chain": ["op-1", "ag-w1"]})
+    res2 = executor.execute("reorder_chain", {"role_id": "developer-1", "desired_chain": ["op-1", "ag-w1"]})
     assert res2["ok"] is True
 
     # edit_route
-    res3 = executor.execute("edit_route", {"role_id": "coder-secondary", "chain": ["op-1", "ag-w2"]})
+    res3 = executor.execute("edit_route", {"role_id": "developer-2", "chain": ["op-1", "ag-w2"]})
     assert res3["ok"] is True
 
     reloaded = load_router_config()
-    assert reloaded.roles["coder-primary"].preferred_chain == ["op-1", "ag-w1"]
-    assert reloaded.roles["coder-secondary"].preferred_chain == ["op-1", "ag-w2"]
+    assert reloaded.roles["developer-1"].preferred_chain == ["op-1", "ag-w1"]
+    assert reloaded.roles["developer-2"].preferred_chain == ["op-1", "ag-w2"]
 
 
 def test_action_executor_set_model_updates_profile_and_role(temp_router_config: Path):
@@ -147,13 +147,13 @@ def test_action_executor_set_model_updates_profile_and_role(temp_router_config:
     res = executor.execute("set_model", {
         "profile_id": "ag-w1",
         "model": "gemini-3.1-pro",
-        "role_id": "coder-primary",
+        "role_id": "developer-1",
     })
     assert res["ok"] is True
 
     reloaded = load_router_config()
     assert reloaded.profiles["ag-w1"].preferred_models[0] == "gemini-3.1-pro"
-    assert reloaded.roles["coder-primary"].default_model == "gemini-3.1-pro"
+    assert reloaded.roles["developer-1"].default_model == "gemini-3.1-pro"
 
 
 def test_web_client_7_views_exact_order_and_no_redundant_views():

tests/test_state_layer_and_event_driven_quota.py

@@ -177,16 +177,16 @@ def test_route_and_agent_delta_events():
     bus.subscribe("AGENT_UPDATED", _on_agent)
 
     try:
-        store.apply_delta_route_changed("coder-primary", "ag-w1", failover_reason="Testing failover")
+        store.apply_delta_route_changed("developer-1", "ag-w1", failover_reason="Testing failover")
         assert len(route_events) >= 1
-        assert route_events[-1]["role_id"] == "coder-primary"
+        assert route_events[-1]["role_id"] == "developer-1"
         assert route_events[-1]["failover_reason"] == "Testing failover"
         assert "generation" in route_events[-1]
         assert "seq" in route_events[-1]
 
         assert len(agent_events) >= 1
-        assert agent_events[-1]["role_id"] == "coder-primary"
-        assert agent_events[-1]["agent"].role_id == "coder-primary"
+        assert agent_events[-1]["role_id"] == "developer-1"
+        assert agent_events[-1]["agent"].role_id == "developer-1"
     finally:
         bus.unsubscribe(EVENT_ROUTING_UPDATED, _on_route)
         bus.unsubscribe("AGENT_UPDATED", _on_agent)

tests/test_state_layer_contract.py

@@ -159,7 +159,7 @@ def test_single_scheduler_refresh_waits_for_quota_before_delta(monkeypatch: pyte
 def test_model_score_rejects_exhausted_pool_and_prefers_more_quota() -> None:
     registry = ModelRegistry.get()
     descriptor = registry.get_model("gemini-2.5-pro")
-    requirements = registry.get_role_requirements("research")
+    requirements = registry.get_role_requirements("researcher")
     assert descriptor is not None
 
     ok_empty, _, reason = registry.evaluate_model_score(

tests/test_ui_claude_grok_connection.py

@@ -1,15 +1,14 @@
-"""End-to-end verification for Grok and Claude connection, assignment, and testing."""
+"""End-to-end verification for Grok and Claude connection, assignment, and testing."""
 
 from __future__ import annotations
 
 from unittest.mock import MagicMock, patch
+
 import pytest
 
-from antigravity_provider.router.auto_assigner import AutoAssigner
-from antigravity_provider.router import action_handler
+from antigravity_provider.router.action_handler import do_test_profile
+from antigravity_provider.router.auto_assigner import AutoAssigner, ensure_profile_in_routing
 from antigravity_provider.router.router_config import RouterConfig, RouterProfileConfig, RolePolicy
-from antigravity_provider.router.ui.add_account_wizard import ensure_profile_in_routing
-from antigravity_provider.router.hermes_hub_app import do_test_profile
 
 
 @pytest.mark.unit
@@ -34,11 +33,10 @@ def test_grok_wizard_definition_and_routing_flow():
     """Verify Grok profile definition and routing pipeline addition."""
     config = RouterConfig(
         profiles={},
-        roles={"coder-primary": RolePolicy(role_name="coder-primary", preferred_chain=[])},
+        roles={"developer-1": RolePolicy(role_name="developer-1", preferred_chain=[])},
     )
     with patch("antigravity_provider.router.auto_assigner.load_router_config", return_value=config), \
-         patch("antigravity_provider.router.auto_assigner.save_router_config", return_value=True), \
-         patch("antigravity_provider.router.ui.add_account_wizard.load_router_config", return_value=config):
+         patch("antigravity_provider.router.auto_assigner.save_router_config", return_value=True):
 
         ok_def, msg_def = AutoAssigner.ensure_profile_definition("grok", "grok-worker-1")
         assert ok_def, f"Definition failed: {msg_def}"
@@ -47,7 +45,7 @@ def test_grok_wizard_definition_and_routing_flow():
 
         ok_route, msg_route = ensure_profile_in_routing("grok-worker-1")
         assert ok_route, f"Routing failed: {msg_route}"
-        assert "grok-worker-1" in config.roles["coder-primary"].preferred_chain
+        assert "grok-worker-1" in config.roles["developer-1"].preferred_chain
 
 
 @pytest.mark.unit
@@ -55,11 +53,10 @@ def test_claude_wizard_definition_and_routing_flow():
     """Verify Claude profile definition and routing pipeline addition."""
     config = RouterConfig(
         profiles={},
-        roles={"orchestrator": RolePolicy(role_name="orchestrator", preferred_chain=[])},
+        roles={"manager": RolePolicy(role_name="manager", preferred_chain=[])},
     )
     with patch("antigravity_provider.router.auto_assigner.load_router_config", return_value=config), \
-         patch("antigravity_provider.router.auto_assigner.save_router_config", return_value=True), \
-         patch("antigravity_provider.router.ui.add_account_wizard.load_router_config", return_value=config):
+         patch("antigravity_provider.router.auto_assigner.save_router_config", return_value=True):
 
         ok_def, msg_def = AutoAssigner.ensure_profile_definition("claude", "claude-orch")
         assert ok_def, f"Definition failed: {msg_def}"
@@ -68,7 +65,7 @@ def test_claude_wizard_definition_and_routing_flow():
 
         ok_route, msg_route = ensure_profile_in_routing("claude-orch")
         assert ok_route, f"Routing failed: {msg_route}"
-        assert "claude-orch" in config.roles["orchestrator"].preferred_chain
+        assert "claude-orch" in config.roles["manager"].preferred_chain
 
 
 @pytest.mark.unit

tests/test_ui_contract_v11.py

@@ -43,8 +43,8 @@ def _profile(plan_code: str = "PRO", plan_source: str = "provider_api") -> Profi
         account_identity="user@example.test",
         provider="antigravity",
         provider_display_name="Google Antigravity",
-        assigned_roles=["coder-primary"],
-        primary_role="coder-primary",
+        assigned_roles=["developer-1"],
+        primary_role="developer-1",
         is_main_account=True,
         is_main_orchestrator=False,
         auth_state="AUTHENTICATED",
@@ -143,7 +143,7 @@ def test_quota_missing_is_not_rendered_as_zero_and_reason_is_visible(ui_root) ->
 @pytest.mark.ui
 def test_agent_quota_and_failover_reason_are_bound_to_their_models(ui_root) -> None:
     agent = AgentViewModel(
-        role_id="coder-primary",
+        role_id="developer-1",
         role_name_ru="Кодер 1",
         role_description_ru="Основной кодер",
         assigned_profile_id="account-2",
@@ -163,7 +163,7 @@ def test_agent_quota_and_failover_reason_are_bound_to_their_models(ui_root) -> N
     )
     team_card = AgentCardWidget(ui_root)
     pipeline = RolePipeline(
-        role_id="coder-primary",
+        role_id="developer-1",
         role_name_ru="Кодер 1",
         default_model="gpt-5",
         max_failover=2,
@@ -214,7 +214,7 @@ def test_agent_quota_and_failover_reason_are_bound_to_their_models(ui_root) -> N
 
 def test_dashboard_agent_quota_measurement_drives_progress_percent() -> None:
     agent = AgentViewModel(
-        role_id="coder-primary",
+        role_id="developer-1",
         role_name_ru="Кодер 1",
         role_description_ru="Основной кодер",
         assigned_profile_id="ag-w1",

tests/test_ui_mockup_redesign.py

@@ -130,7 +130,7 @@ def test_restored_account_buttons_invoke_each_action(ui_root) -> None:
 @pytest.mark.ui
 def test_assign_role_error_stays_visible_in_open_modal(ui_root, monkeypatch) -> None:
     policy = SimpleNamespace(preferred_chain=[])
-    monkeypatch.setattr(app_module, "load_router_config", lambda: SimpleNamespace(roles={"orchestrator": policy}))
+    monkeypatch.setattr(app_module, "load_router_config", lambda: SimpleNamespace(roles={"manager": policy}))
     monkeypatch.setattr(
         app_module.AutoAssigner,
         "assign_profile_to_role",

tests/test_ui_phase2_6.py

@@ -37,8 +37,8 @@ def _profile(index: int) -> ProfileViewModel:
         account_identity=f"user-{index}@example.test",
         provider="antigravity",
         provider_display_name="Google Antigravity",
-        assigned_roles=["coder" if index else "orchestrator"],
-        primary_role="coder" if index else "orchestrator",
+        assigned_roles=["coder" if index else "manager"],
+        primary_role="coder" if index else "manager",
         is_main_account=index == 0,
         is_main_orchestrator=index == 0,
         auth_state="AUTHENTICATED",
@@ -117,6 +117,7 @@ def test_identity_priority_and_plan_badge_suppression() -> None:
 
 
 @pytest.mark.ui
+@pytest.mark.timeout(90)
 def test_fifty_accounts_update_one_quota_without_rebuilding_other_cards(ui_root) -> None:
     view = AccountsView(ui_root)
     try:

tests/test_ui_routing_graph.py

@@ -29,12 +29,12 @@ def _config():
         "coder": SimpleNamespace(provider="antigravity", preferred_models=["gemini"]),
     }
     roles = {
-        "orchestrator": SimpleNamespace(preferred_chain=["orch"]),
-        "coder-primary": SimpleNamespace(preferred_chain=["coder"]),
-        "coder-secondary": SimpleNamespace(preferred_chain=["coder"]),
-        "reviewer": SimpleNamespace(preferred_chain=["coder"]),
-        "research": SimpleNamespace(preferred_chain=["coder"]),
-        "fast": SimpleNamespace(preferred_chain=["coder"]),
+        "manager": SimpleNamespace(preferred_chain=["orch"]),
+        "developer-1": SimpleNamespace(preferred_chain=["coder"]),
+        "developer-2": SimpleNamespace(preferred_chain=["coder"]),
+        "code-reviewer": SimpleNamespace(preferred_chain=["coder"]),
+        "researcher": SimpleNamespace(preferred_chain=["coder"]),
+        "tester": SimpleNamespace(preferred_chain=["coder"]),
     }
     return SimpleNamespace(roles=roles, profiles=profiles)
 
@@ -83,16 +83,16 @@ def test_graph_layout_zoom_and_viewport_survive_restart(tmp_path):
 
 def test_validation_finds_cycle_unreachable_and_missing_profile():
     config = _config()
-    config.roles["reviewer"].preferred_chain = ["ghost"]
+    config.roles["code-reviewer"].preferred_chain = ["ghost"]
     graph = RoutingGraph(
         nodes=[
-            GraphNode("orchestrator", 0, 0),
-            GraphNode("coder-primary", 1, 0),
-            GraphNode("reviewer", 2, 0),
+            GraphNode("manager", 0, 0),
+            GraphNode("developer-1", 1, 0),
+            GraphNode("code-reviewer", 2, 0),
         ],
         edges=[
-            GraphEdge("orchestrator", "coder-primary"),
-            GraphEdge("coder-primary", "orchestrator"),
+            GraphEdge("manager", "developer-1"),
+            GraphEdge("developer-1", "manager"),
         ],
     )
     codes = {issue.code for issue in validate_graph(graph, config)}
@@ -108,16 +108,16 @@ def test_profile_edge_updates_yaml_via_auto_assigner(monkeypatch, tmp_path):
         lambda profile, role, is_primary: calls.append((profile, role, is_primary)) or (True, "ok"),
     )
     controller = RoutingGraphController(RoutingGraphStore(tmp_path / "graph.json"))
-    ok, _message = controller.add_edge("orchestrator", "coder-primary", "FALLBACK", "orch")
+    ok, _message = controller.add_edge("manager", "developer-1", "FALLBACK", "orch")
     assert ok
-    assert calls == [("orch", "coder-primary", False)]
+    assert calls == [("orch", "developer-1", False)]
 
 
 def test_undo_redo_and_dirty_state(monkeypatch, tmp_path):
     monkeypatch.setattr(graph_module, "load_router_config", _config)
     controller = RoutingGraphController(RoutingGraphStore(tmp_path / "graph.json"))
     original = controller.graph.nodes[0].x
-    controller.move_node("orchestrator", original + 100, 50)
+    controller.move_node("manager", original + 100, 50)
     assert controller.dirty
     assert controller.undo()
     assert controller.graph.nodes[0].x == original
@@ -133,18 +133,20 @@ def test_graph_store_handles_twenty_nodes(tmp_path):
 
 
 def test_wizard_keeps_existing_chain_rank_and_assigns_missing_slot(monkeypatch):
+    from antigravity_provider.router import auto_assigner as assigner_module
+
     config = _config()
     calls = []
-    monkeypatch.setattr(wizard_module, "load_router_config", lambda: config)
+    monkeypatch.setattr(assigner_module, "load_router_config", lambda: config)
     monkeypatch.setattr(
-        wizard_module.AutoAssigner,
+        assigner_module.AutoAssigner,
         "assign_profile_to_role",
         lambda profile, role, is_primary: calls.append((profile, role, is_primary)) or (True, "ok"),
     )
     assert wizard_module.ensure_profile_in_routing("orch")[0]
     assert calls == []
     monkeypatch.setattr(
-        wizard_module.AutoAssigner,
+        assigner_module.AutoAssigner,
         "get_display_name_and_role",
         lambda _profile: ("Новый кодер", "coder", "fallback"),
     )
@@ -255,8 +257,8 @@ def test_role_chain_order_and_removal_persist_through_auto_assigner(monkeypatch)
     config = SimpleNamespace(
         profiles={key: SimpleNamespace() for key in ("a", "b", "c")},
         roles={
-            "orchestrator": SimpleNamespace(preferred_chain=["a", "b", "c"]),
-            "reviewer": SimpleNamespace(preferred_chain=["b"]),
+            "manager": SimpleNamespace(preferred_chain=["a", "b", "c"]),
+            "code-reviewer": SimpleNamespace(preferred_chain=["b"]),
         },
     )
     calls = []
@@ -279,11 +281,11 @@ def test_role_chain_order_and_removal_persist_through_auto_assigner(monkeypatch)
     monkeypatch.setattr(team_module, "load_router_config", lambda: config)
     monkeypatch.setattr(team_module.AutoAssigner, "assign_profile_to_role", assign)
 
-    ok, _message = team_module.persist_role_chain("orchestrator", ["c", "a"])
+    ok, _message = team_module.persist_role_chain("manager", ["c", "a"])
 
     assert ok
-    assert config.roles["orchestrator"].preferred_chain == ["c", "a"]
-    assert config.roles["reviewer"].preferred_chain == ["b"]
+    assert config.roles["manager"].preferred_chain == ["c", "a"]
+    assert config.roles["code-reviewer"].preferred_chain == ["b"]
     assert ("b", "spare", False) in calls
 
 

tests/test_unified_health.py

@@ -96,7 +96,7 @@ def test_main_account_vs_orchestrator_separation():
     service.scan_all()
     agents = service.get_agent_view_models()
 
-    orch_agent = next((a for a in agents if a.role_id == "orchestrator"), None)
+    orch_agent = next((a for a in agents if a.role_id == "manager"), None)
     assert orch_agent is not None
     assert orch_agent.is_main_orchestrator is True
 

tests/test_web_parity_a21.py

@@ -123,14 +123,12 @@ def test_web_client_html_and_js_7_views_parity():
 
     # Routing view elements
     assert "renderRoutingView" in app_js
-    assert "routing-pipelines-container" in index_html
-    assert "handleNodeDragStart" in app_js
-    assert "handleNodeDrop" in app_js
-    assert "handleNodeModelChange" in app_js
+    assert "routing-roles-container" in index_html or "routing-pipelines-container" in index_html
+    assert "setupDragAndDrop" in app_js or "handleNodeDragStart" in app_js
 
     # Overview view elements
     assert "renderOverviewView" in app_js
-    assert "overview-route-diagram" in index_html
+    assert "workflow-canvas" in index_html
 
     # Analytics view elements
     assert "analytics-total-calls" in index_html

tests/test_workflow_service_a30.py

@@ -0,0 +1,169 @@
+from __future__ import annotations
+
+import json
+
+import pytest
+
+from antigravity_provider.router.router_config import (
+    RolePolicy,
+    RouterConfig,
+    RouterProfileConfig,
+    save_router_config,
+)
+from antigravity_provider.router.workflow_service import WorkflowService
+
+
+@pytest.fixture
+def workflow_service(tmp_path, monkeypatch):
+    monkeypatch.setenv("HERMES_HOME", str(tmp_path))
+    config = RouterConfig(
+        profiles={
+            "account-a": RouterProfileConfig(
+                profile_id="account-a",
+                provider="openai-codex",
+                preferred_models=["model-real-from-config"],
+            )
+        },
+        roles={"developer": RolePolicy(role_name="developer", preferred_chain=["account-a"])},
+    )
+    save_router_config(config)
+    return WorkflowService(tmp_path / "workflow_state.json")
+
+
+def test_router_roles_migrate_to_agents_and_create_real_files(workflow_service, tmp_path):
+    snapshot = workflow_service.snapshot()
+    assert "developer" in [agent["id"] for agent in snapshot["agents"]]
+    agent = next(item for item in snapshot["agents"] if item["id"] == "developer")
+    assert agent["agent_file"] == "agents/developer.md"
+    assert agent["agent_file_exists"] is True
+    assert (tmp_path / "agents" / "developer.md").is_file()
+    assert agent["execution_config"]["account"] == "account-a"
+    assert agent["execution_config"]["model"] == "model-real-from-config"
+
+
+def test_create_update_file_and_restart_persistence(workflow_service, tmp_path):
+    created = workflow_service.create_agent({
+        "name": "Security Reviewer",
+        "role": "security-reviewer",
+        "account": "account-a",
+        "model": "model-real-from-config",
+        "description": "Проверяет безопасность",
+    })
+    workflow_service.save_agent_file(created.id, "# Security\n\nOnly measured facts.")
+    workflow_service.update_agent(created.id, {"timeout": 91, "position": {"x": 33, "y": 44}})
+
+    restarted = WorkflowService(tmp_path / "workflow_state.json")
+    item = next(agent for agent in restarted.snapshot()["agents"] if agent["id"] == created.id)
+    assert item["timeout"] == 91
+    assert item["position"] == {"x": 33, "y": 44}
+    assert restarted.read_agent_file(created.id)["content"].endswith("Only measured facts.")
+
+
+def test_delete_requires_explicit_confirmation_when_referenced(workflow_service):
+    workflow_service.create_agent({"name": "Test Reviewer", "role": "test-reviewer", "account": "account-a"})
+    workflow_service.save_workflow({
+        "start_agent_id": "developer",
+        "max_iterations": 3,
+        "edges": [{"id": "review", "source": "developer", "target": "test-reviewer", "condition": "SUCCESS"}],
+    })
+    warning = workflow_service.delete_agent("test-reviewer")
+    assert warning["confirmation_required"] is True
+    assert warning["consequences"]["workflow_edges"] == ["review"]
+    result = workflow_service.delete_agent("test-reviewer", force=True)
+    assert result["deleted"] is True
+    assert workflow_service.workflow.edges == []
+
+
+def test_cycles_are_valid_and_iteration_limit_is_persisted(workflow_service):
+    workflow_service.create_agent({"name": "Test Reviewer", "role": "test-reviewer", "account": "account-a"})
+    definition = workflow_service.save_workflow({
+        "start_agent_id": "developer",
+        "max_iterations": 2,
+        "edges": [
+            {"source": "developer", "target": "test-reviewer", "condition": "SUCCESS"},
+            {"source": "test-reviewer", "target": "developer", "condition": "REVIEW_FAILED"},
+        ],
+    })
+    assert definition.max_iterations == 2
+    assert definition.edges[1].condition == "REVIEW_FAILED"
+    payload = json.loads(workflow_service.state_path.read_text(encoding="utf-8"))
+    assert payload["workflow"]["max_iterations"] == 2
+
+
+def test_invalid_edge_and_unknown_model_are_rejected(workflow_service):
+    with pytest.raises(ValueError, match="отсутствующего агента"):
+        workflow_service.save_workflow({"edges": [{"source": "developer", "target": "missing"}]})
+    with pytest.raises(ValueError, match="не доступна"):
+        workflow_service.update_agent("developer", {"account": "account-a", "model": "invented-model"})
+
+
+def test_interrupted_run_is_reported_not_silently_completed(workflow_service, tmp_path):
+    workflow_service.run.update({"id": "run-1", "status": "running", "current_agent_id": "developer"})
+    workflow_service._save()
+    restarted = WorkflowService(tmp_path / "workflow_state.json")
+    assert restarted.run["status"] == "interrupted"
+    assert "перезапуском" in restarted.run["error"]
+    assert restarted.events[-1].type == "WORKFLOW_INTERRUPTED"
+
+
+def test_live_cycle_stops_with_explicit_iteration_limit_event(workflow_service, monkeypatch):
+    workflow_service.create_agent({"name": "Loop Reviewer", "role": "loop-reviewer", "account": "account-a"})
+    workflow_service.save_workflow({
+        "start_agent_id": "developer",
+        "max_iterations": 2,
+        "edges": [
+            {"source": "developer", "target": "loop-reviewer", "condition": "SUCCESS"},
+            {"source": "loop-reviewer", "target": "developer", "condition": "REVIEW_FAILED"},
+        ],
+    })
+
+    class FakeEngine:
+        def reload_config(self):
+            return None
+
+        def route_request(self, request, role=None, session_id=None):
+            status = "REVIEW_FAILED" if role == "loop-reviewer" else "SUCCESS"
+            return {
+                "choices": [{"message": {"content": status}}],
+                "router_metadata": {
+                    "provider": "measured-provider",
+                    "profile_id": "account-a",
+                    "selection_trace": {"selected_model": "model-real-from-config"},
+                },
+            }
+
+    monkeypatch.setattr("antigravity_provider.router.router_engine.get_router_engine", lambda: FakeEngine())
+    workflow_service.start("Проверить реальный цикл")
+    thread = workflow_service._thread
+    assert thread is not None
+    thread.join(timeout=3)
+    assert not thread.is_alive(), "Workflow execution thread leaked after join"
+
+    assert workflow_service.run["status"] == "failed"
+    assert workflow_service.run["error"] == "Достигнут предел итераций: 2"
+    assert any(event.type == "WORKFLOW_MAX_ITERATIONS" for event in workflow_service.events)
+
+
+def test_provider_error_text_reaches_run_and_events(workflow_service, monkeypatch):
+    provider_text = "Provider Error: authentication token missing for account-a"
+
+    class ErrorEngine:
+        def reload_config(self):
+            return None
+
+        def route_request(self, request, role=None, session_id=None):
+            return {"choices": [{"message": {"content": f"ERROR\n{provider_text}"}}]}
+
+    monkeypatch.setattr("antigravity_provider.router.router_engine.get_router_engine", lambda: ErrorEngine())
+    workflow_service.workflow.start_agent_id = "developer"
+    workflow_service.workflow.edges = []
+    workflow_service.start("Проверить ошибку")
+    thread = workflow_service._thread
+    assert thread is not None
+    thread.join(timeout=3)
+    assert not thread.is_alive(), "Workflow execution thread leaked after join"
+
+    assert workflow_service.run["status"] == "failed"
+    assert provider_text in workflow_service.run["error"]
+    assert any(provider_text in (event.error or "") for event in workflow_service.events)
+    assert workflow_service.snapshot()["agents"][0]["runtime_state"] == "error"