@team-agent/installer 0.2.3 → 0.2.5

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@team-agent/installer",
-  "version": "0.2.3",
+  "version": "0.2.5",
   "description": "npx installer for Team Agent",
   "keywords": [
     "codex",

package/src/team_agent/abnormal_track.py

@@ -0,0 +1,253 @@
+"""Provider-neutral abnormal-state track (Gap 32 §4).
+
+Reads structured fault records + process identity; never reads a screen and
+never names a provider. Catch-bias for structured error/failed-class records
+(C9), dedup by (signature, turn) (C8), and coordinator-independent whole-team
+disappearance with clean-shutdown vs unexpected distinction (C10).
+"""
+
+from __future__ import annotations
+
+from typing import Any
+
+
+def process_abnormal_records(
+    records: list[dict[str, Any]],
+    *,
+    registry: Any,
+    notification_state: dict[str, Any] | None,
+    event_sink: Any = None,
+) -> dict[str, Any]:
+    """Classify raw provider session records that may carry faults.
+
+    ``registry`` carries the provider whose records these are (``{"provider":
+    name}``) or a full registry mapping. Records are turned into structured
+    fault facts by the provider reader (so this module names no provider), then
+    catch-biased + deduped by (signature, turn).
+    """
+    from team_agent.provider_state import read_fault_facts
+    from team_agent.provider_state.registry import get_provider_registry
+
+    state = dict(notification_state or {})
+    seen = set(state.get("seen") or [])
+    notifications: list[dict[str, Any]] = []
+    discovery_log: list[dict[str, Any]] = []
+    diagnostics: list[dict[str, Any]] = []
+
+    provider = _provider_of(registry)
+    white, black = _lists_for(provider, registry, get_provider_registry)
+
+    faults = read_fault_facts(provider, records or []) if provider else []
+    if not faults and records:
+        # Records that produced no structured fault fact are not default-notify
+        # candidates (C9): arbitrary unrecognized lines become diagnostics only.
+        diagnostics.append({"kind": "no_structured_fault", "count": len(records)})
+
+    for fact in faults:
+        signature = str(fact.get("signature") or fact.get("reason") or "fault")
+        turn_id = fact.get("turn_id")
+        text = " ".join(str(x) for x in (signature, fact.get("reason"), _raw_message(fact)) if x).lower()
+        decision = _classify(text, signature, white, black)
+        discovery_log.append({
+            "signature": signature,
+            "turn_id": turn_id,
+            "decision": decision,
+            "kind": fact.get("kind"),
+            "provider": provider,
+        })
+        if decision == "skip":
+            continue
+        # C8: dedup by (signature, turn_id) — a retry loop in the SAME turn folds
+        # to one notify. But a MISSING turn_id must not collapse distinct errors
+        # into one global bucket: discriminate by a per-record content fingerprint
+        # so genuinely different faults each notify (identical duplicates still fold).
+        bucket = turn_id if turn_id is not None else f"norow:{_record_fingerprint(fact)}"
+        dedupe_key = (signature, bucket)
+        key = f"{signature}\x00{bucket}"
+        if key in seen:
+            continue
+        seen.add(key)
+        notifications.append({
+            "signature": signature,
+            "turn_id": turn_id,
+            "dedupe_key": dedupe_key,
+            "state": "blocked_on_human" if fact.get("kind") == "approval" else "abnormal",
+            "decision": decision,
+            "provider": provider,
+            "raw": fact.get("raw", fact),
+            "raw_record": fact.get("raw", fact),
+        })
+        _emit(event_sink, "abnormal.notify", signature=signature, turn_id=turn_id, decision=decision)
+
+    state["seen"] = sorted(seen)
+    return {
+        "notifications": notifications,
+        "discovery_log": discovery_log,
+        "diagnostics": diagnostics,
+        "notification_state": state,
+    }
+
+
+def detect_whole_team_gone(
+    snapshot: dict[str, Any],
+    *,
+    marker_store: Any,
+    event_sink: Any = None,
+) -> dict[str, Any]:
+    """Coordinator-independent whole-team-gone detection (C10/C13).
+
+    Does not require the coordinator to be alive. The whole team is gone when the
+    coordinator, the leader, every provider process, and every session are all
+    absent. Clean shutdown / restart-in-progress (flagged in the snapshot) are
+    silent; an unexpected disappearance records a durable marker and defers user
+    escalation to the next leader command.
+    """
+    coordinator = snapshot.get("coordinator") or {}
+    leader = snapshot.get("leader") or {}
+    provider_processes = snapshot.get("provider_processes")
+    if provider_processes is None:
+        provider_processes = snapshot.get("nodes") or snapshot.get("agents") or []
+    tmux_sessions = snapshot.get("tmux_sessions") or []
+
+    coord_alive = _alive(coordinator)
+    leader_alive = _alive(leader)
+    any_worker_alive = any(_alive(p) for p in provider_processes)
+    sessions_present = bool(tmux_sessions)
+
+    whole_gone = not (coord_alive or leader_alive or any_worker_alive or sessions_present)
+
+    if not whole_gone:
+        return {
+            "state": "alive",
+            "whole_team_gone": False,
+            "classification": "alive",
+            "notify": False,
+            "escalate_user_on_next_leader_command": False,
+            "marker_written": False,
+        }
+
+    if snapshot.get("clean_shutdown"):
+        return _silent_gone("clean_shutdown")
+    if snapshot.get("restart_in_progress"):
+        return _silent_gone("restart_in_progress")
+
+    # Unexpected disappearance (闪退): durable marker + deferred escalation.
+    marker_written = _marker_set(marker_store, "whole_team_gone", {
+        "classification": "unexpected_exit",
+        "provider_processes": len(provider_processes),
+    })
+    _emit(event_sink, "abnormal.whole_team_gone", classification="unexpected_exit")
+    return {
+        "state": "whole_team_gone",
+        "whole_team_gone": True,
+        "classification": "unexpected_exit",
+        "notify": True,
+        "escalate_user_on_next_leader_command": True,
+        "marker_written": bool(marker_written),
+    }
+
+
+def _silent_gone(classification: str) -> dict[str, Any]:
+    return {
+        "state": classification,
+        "whole_team_gone": True,
+        "classification": classification,
+        "notify": False,
+        "escalate_user_on_next_leader_command": False,
+        "marker_written": False,
+    }
+
+
+def _alive(entry: Any) -> bool:
+    from team_agent.provider_state.common import process_is_live
+
+    if isinstance(entry, dict):
+        if "alive" in entry:
+            return entry.get("alive") is True
+        if "process" in entry:
+            ok, _r, _d = process_is_live(entry.get("process"))
+            return ok
+        ok, _r, _d = process_is_live(entry)
+        return ok
+    return bool(entry)
+
+
+def _provider_of(registry: Any) -> str | None:
+    if isinstance(registry, dict):
+        if isinstance(registry.get("provider"), str):
+            return registry.get("provider")
+        if isinstance(registry.get("kind"), str):
+            return registry.get("kind")
+    return None
+
+
+def _lists_for(provider: str | None, registry: Any, get_provider_registry: Any) -> tuple[list[str], list[str]]:
+    entry: Any = None
+    if isinstance(registry, dict) and ("error_whitelist" in registry or "error_blacklist" in registry):
+        entry = registry
+    elif provider is not None:
+        entry = get_provider_registry(provider)
+    if not isinstance(entry, dict):
+        return [], []
+    lists = entry.get("error_lists") if isinstance(entry.get("error_lists"), dict) else {}
+    white = [str(x).lower() for x in (lists.get("whitelist") or entry.get("error_whitelist") or [])]
+    black = [str(x).lower() for x in (lists.get("blacklist") or entry.get("error_blacklist") or [])]
+    return white, black
+
+
+def _classify(text: str, signature: str, white: list[str], black: list[str]) -> str:
+    sig = signature.lower()
+    if any(w and (w in text or w in sig) for w in white):
+        return "skip"  # whitelist > blacklist > default
+    if any(b and (b in text or b in sig) for b in black):
+        return "notify_blacklist"
+    return "notify_default"  # C9 catch-bias for structured faults
+
+
+def _record_fingerprint(fact: dict[str, Any]) -> str:
+    import hashlib
+    import json
+
+    raw = fact.get("raw", fact)
+    try:
+        blob = json.dumps(raw, sort_keys=True, default=str)
+    except (TypeError, ValueError):
+        blob = repr(raw)
+    return hashlib.sha256(blob.encode("utf-8", errors="ignore")).hexdigest()[:16]
+
+
+def _raw_message(fact: dict[str, Any]) -> str:
+    raw = fact.get("raw")
+    if isinstance(raw, dict):
+        return str(raw.get("message") or "")
+    return ""
+
+
+def _marker_set(marker_store: Any, name: str, value: Any) -> bool:
+    if marker_store is None:
+        return False
+    if isinstance(marker_store, dict):
+        marker_store[name] = value
+        return True
+    setter = getattr(marker_store, "set", None) or getattr(marker_store, "write", None)
+    if callable(setter):
+        try:
+            setter(name, value)
+            return True
+        except Exception:
+            return False
+    return False
+
+
+def _emit(event_sink: Any, name: str, **fields: Any) -> None:
+    if event_sink is None:
+        return
+    try:
+        event_sink(name, fields)
+    except TypeError:
+        try:
+            event_sink({"event": name, **fields})
+        except Exception:
+            pass
+    except Exception:
+        pass

package/src/team_agent/cli/commands.py

@@ -224,6 +224,20 @@ def cmd_doctor(args: argparse.Namespace) -> dict[str, Any] | str:
         if gate != "orphans":
             raise TeamAgentError(f"unknown doctor gate: {gate}")
         return orphan_gate(fix=bool(getattr(args, "fix", False)), confirm=bool(getattr(args, "confirm", False)))
+    from team_agent.message_store.schema import SCHEMA_VERSION
+    from team_agent.message_store.schema_migration import fix_schema_layout, schema_diagnosis
+    if getattr(args, "fix_schema", False) is True:
+        return fix_schema_layout(Path(args.workspace).resolve(), schema_version=SCHEMA_VERSION)
+    schema = schema_diagnosis(Path(args.workspace).resolve(), schema_version=SCHEMA_VERSION)
+    if schema.get("layout_diffs"):
+        return {
+            "ok": True,
+            "schema": schema,
+            "coordinator": {
+                "schema_ok": False,
+                "schema_error": "team.db physical layout drift detected",
+            },
+        }
     if getattr(args, "cleanup_orphans", False):
         from team_agent.diagnose.orphan_cleanup import cleanup_orphan_coordinators, format_cleanup_orphans
         result = cleanup_orphan_coordinators(confirm=bool(getattr(args, "confirm", False)))
@@ -231,7 +245,9 @@ def cmd_doctor(args: argparse.Namespace) -> dict[str, Any] | str:
             return result
         return format_cleanup_orphans(result)
     spec = Path(args.spec).resolve() if args.spec else None
-    return runtime.doctor(spec)
+    result = runtime.doctor(spec)
+    result["schema"] = schema
+    return result
 
 
 def _format_status_summary(data: dict[str, Any]) -> str:

package/src/team_agent/cli/parser.py

@@ -54,7 +54,6 @@ from team_agent.cli.commands import (
     cmd_advanced,
     cmd_install_skill,
     cmd_run_overnight,
-
 )
 from team_agent.cli.e2e import cmd_e2e
 from team_agent.cli.helpers import (
@@ -65,7 +64,6 @@ from team_agent.cli.helpers import (
     emit,
 )
 
-
 SEND_ORDER_HINT = (
     "options must appear before target/message. Use: "
     "team-agent send --task <task_id> --json \"<message>\" or "
@@ -317,8 +315,10 @@ def main(argv: list[str] | None = None) -> None:
 
     p = sub.add_parser("doctor", help="Check local dependencies, providers, auth hints, tmux, and MCP")
     p.add_argument("spec", nargs="?")
+    p.add_argument("--workspace", default=".", help="Workspace whose team.db schema should be diagnosed")
     p.add_argument("--gate", choices=["orphans"], help="Run a CI-friendly doctor gate")
     p.add_argument("--fix", action="store_true", help="With --gate orphans: apply the gate fix")
+    p.add_argument("--fix-schema", action="store_true", help="Rebuild drifted team.db table layouts after writing a backup")
     p.add_argument(
         "--cleanup-orphans",
         action="store_true",

package/src/team_agent/compiler.py

@@ -93,7 +93,7 @@ def compile_team(team_dir: Path, out_path: Path | None = None) -> dict[str, Any]
         },
         "runtime": {
             "backend": "tmux",
-            "display_backend": str(team_meta.get("display_backend") or "ghostty_window"),
+            "display_backend": str(team_meta.get("display_backend") or "adaptive"),
             "session_name": str(team_meta.get("session_name") or f"team-{_slug(team_name)}"),
             "auto_launch": True,
             "require_user_approval_before_launch": True,

package/src/team_agent/coordinator/lifecycle.py

@@ -262,8 +262,10 @@ def coordinator_tick(workspace: Path) -> dict[str, Any]:
     )
     from team_agent.messaging.idle_alerts import (
         detect_cross_worker_deadlocks,
-        detect_idle_fallbacks,
     )
+    from team_agent.idle_predicate import evaluate_takeover_reminder
+    from team_agent.idle_takeover_wiring import build_idle_nodes, push_idle_reminder, IDLE_DEBOUNCE_SECONDS
+    import time as _time
     from team_agent.messaging.activity_detector import detect_compaction_degradation
     from team_agent.messaging.leader_api_errors import detect_leader_api_errors
     from team_agent.messaging.session_drift import detect_session_drift
@@ -283,7 +285,23 @@ def coordinator_tick(workspace: Path) -> dict[str, Any]:
     delivered = _deliver_pending_messages(workspace, state, event_log)
     fired = _fire_due_scheduled_events(workspace, store, event_log)
     stuck = _detect_stuck_agents(workspace, state, store, event_log)
-    idle_alerts = detect_idle_fallbacks(workspace, state, store, event_log)
+    # Gap 32: the take-over reminder is driven by file-fact turn-state via the
+    # idle_takeover predicate (the legacy screen-scrape obligation path is retired).
+    _coord_meta = state.setdefault("coordinator", {})
+    idle_eval = evaluate_takeover_reminder(
+        build_idle_nodes(state),
+        monitor_state=_coord_meta.get("idle_takeover_monitor"),
+        now_monotonic=_time.monotonic(),
+        debounce_seconds=IDLE_DEBOUNCE_SECONDS,
+    )
+    _coord_meta["idle_takeover_monitor"] = idle_eval.get("monitor_state")
+    push_idle_reminder(workspace, state, event_log, idle_eval)
+    idle_alerts = (
+        [{"alert_type": "idle_takeover", "message": idle_eval.get("message"),
+          "reason": idle_eval.get("reason"), "interrupted": idle_eval.get("interrupted_nodes")}]
+        if idle_eval.get("should_ping")
+        else []
+    )
     deadlock_alerts = detect_cross_worker_deadlocks(workspace, state, store, event_log)
     compaction_results: list[dict[str, Any]] = []
     for agent_id, agent_state in state.get("agents", {}).items():

package/src/team_agent/display/__init__.py

@@ -1,9 +1,27 @@
 from __future__ import annotations
 
+from team_agent.display.adaptive import (
+    ADAPTIVE_BLOCK_REASONS,
+    adaptive_blocked,
+    close_adaptive_display,
+    close_adaptive_windows,
+    open_adaptive_display,
+    prepare_adaptive_windows,
+    probe_display_capabilities,
+)
+from team_agent.display.backend import (
+    ADAPTIVE_DISPLAY_BACKEND,
+    DISPLAY_BACKENDS_WITH_WORKER_VIEWS,
+    GHOSTTY_DISPLAY_BACKENDS,
+    VALID_DISPLAY_BACKENDS,
+    resolve_display_backend,
+)
 from team_agent.display.close import (
+    close_adaptive_display,
     close_ghostty_display,
     close_ghostty_workspace,
     close_ghostty_workspace_slot,
+    close_team_display_backends,
 )
 from team_agent.display.ghostty import (
     ghostty_app_exists,
@@ -34,10 +52,19 @@ from team_agent.display.workspace import (
 )
 
 __all__ = [
+    "ADAPTIVE_BLOCK_REASONS",
+    "ADAPTIVE_DISPLAY_BACKEND",
+    "DISPLAY_BACKENDS_WITH_WORKER_VIEWS",
     "GHOSTTY_WORKSPACE_PANES_PER_WINDOW",
+    "GHOSTTY_DISPLAY_BACKENDS",
+    "VALID_DISPLAY_BACKENDS",
+    "adaptive_blocked",
+    "close_adaptive_display",
+    "close_adaptive_windows",
     "close_ghostty_display",
     "close_ghostty_workspace",
     "close_ghostty_workspace_slot",
+    "close_team_display_backends",
     "ghostty_app_exists",
     "ghostty_attach_args",
     "ghostty_command",
@@ -51,11 +78,15 @@ __all__ = [
     "ghostty_workspace_window_name",
     "kill_ghostty_workspace_linked_sessions",
     "open_ghostty_worker_window",
+    "open_adaptive_display",
     "open_ghostty_workspace",
     "open_ghostty_workspace_agent_display",
     "open_worker_displays",
+    "prepare_adaptive_windows",
     "prepare_ghostty_display_session",
     "prepare_ghostty_workspace_aggregator",
     "prepare_ghostty_workspace_linked_sessions",
+    "probe_display_capabilities",
+    "resolve_display_backend",
     "set_ghostty_workspace_pane_title",
 ]