Merge branch 'master' into feat/issue-658-mcp-restart-coordinator

Child of umbrella #655 (governed MCP restart coordination); builds on the
#657 restart-path inventory. Adds a central coordinator that evaluates live
control-plane state before a restart and returns a blast-radius impact
preview, so operators and the web console (#642/#652) can see what a restart
would disrupt before concurrent LLM work is destroyed.

Changes
- restart_coordinator.py (new) — pure classification: inventory -> impact
  report DTO (RestartImpactReport/SessionImpact/LeaseImpact). Verdicts:
  safe / unsafe / override. Never restarts anything; fails closed on an
  incomplete inventory.
- control_plane_db.py — additive ControlPlaneDB.list_sessions() read-only
  session inventory (the process-level unit a restart kills).
- gitea_mcp_server.py — new dry-run MCP tool gitea_request_mcp_restart:
  gathers sessions/leases/terminal-lock from the #613 DB, calls the
  coordinator, returns the report. Override authority is read from the
  environment, never self-asserted (#630/#710 F1 pattern). Apply is gated
  by a later drain proof (non-goal here).
- docs/mcp-restart-coordinator.md + docs/mcp-restart-impact-sample.json — doc
  and a real dry-run sample report.
- docs/mcp-tool-inventory.md — register the new tool (inventory sync).
- tests/test_restart_coordinator.py (new) — 15 tests: multi-session fixtures,
  deny-when-critical-section-open, fail-closed deny, override, terminal lock,
  stale heartbeat, JSON-serializable DTO, list_sessions.

Tests: pytest tests/test_restart_coordinator.py -> 15 passed. Full suite:
13 failed / 4753 passed; all 13 reproduce identically on clean master
@ef14622 (0 regressions). The residual test_issue_781 doc-registry failure
is a pre-existing baseline gap for gitea_rebind_dirty_same_claimant_author_session
(merged #864, undocumented on master) — out of scope for #658.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

control_plane_db.py

@@ -599,6 +599,35 @@ class ControlPlaneDB:
                 (_ts(), session_id),
             )
 
+    def list_sessions(
+        self,
+        *,
+        statuses: Sequence[str] | None = None,
+        limit: int = 500,
+    ) -> list[dict[str, Any]]:
+        """List session rows for restart / impact analysis (#658).
+
+        Read-only. Sessions are the process-level unit an MCP restart
+        disrupts, so the restart coordinator inventories them to compute blast
+        radius. Optional ``statuses`` filter (e.g. ``('active',)``) narrows to
+        live rows. Never returns secrets — only operational metadata.
+        """
+        clauses: list[str] = []
+        params: list[Any] = []
+        if statuses:
+            placeholders = ", ".join("?" for _ in statuses)
+            clauses.append(f"status IN ({placeholders})")
+            params.extend(statuses)
+        where = ("WHERE " + " AND ".join(clauses)) if clauses else ""
+        sql = (
+            f"SELECT * FROM sessions {where} "
+            "ORDER BY last_heartbeat_at DESC LIMIT ?"
+        )
+        params.append(max(1, int(limit)))
+        with self._tx(immediate=False) as conn:
+            rows = conn.execute(sql, params).fetchall()
+            return [dict(r) for r in rows]
+
     # ── work items ────────────────────────────────────────────────────────
 
     def upsert_work_item(

docs/mcp-restart-coordinator.md

@@ -0,0 +1,95 @@
+# MCP restart coordinator and impact analysis (#658)
+
+Before any sanctioned MCP restart, a central coordinator evaluates the live
+control-plane state and produces an **impact preview** so operators and the web
+console (#642 / #652) can see the blast radius *before* concurrent LLM work is
+disrupted. Uncoordinated restarts destroy in-flight author/reviewer/merger work
+and give operators no way to see what they are about to break.
+
+This lands the coordinator + impact DTO + a dry-run MCP tool. It is the single
+sanctioned entry point for restart evaluation post-#657 (which inventoried the
+restart/reload/kill paths). The **mutative apply** path — actually performing a
+restart — is a later child gated by a drain proof and is explicitly out of
+scope here.
+
+## Components
+
+| Piece | Where | Responsibility |
+|-------|-------|----------------|
+| `restart_coordinator.evaluate_restart_impact` | `restart_coordinator.py` | Pure classification: inventory → impact report DTO. No I/O, no restart. |
+| `RestartImpactReport` / `SessionImpact` / `LeaseImpact` | `restart_coordinator.py` | Console-facing DTO (`.as_dict()` is JSON-serializable). |
+| `ControlPlaneDB.list_sessions` | `control_plane_db.py` | Read-only session inventory (the process-level unit a restart kills). |
+| `gitea_request_mcp_restart` | `gitea_mcp_server.py` | MCP tool: gathers inventory from the #613 DB, calls the coordinator, returns the report. Dry-run only. |
+
+## Dimensions evaluated
+
+The coordinator classifies the inventory across the dimensions #658 requires:
+
+- **Sessions** — every active MCP session; a restart terminates all of them.
+  Liveness = `status == active` **and** the owner pid is alive **and** the
+  heartbeat is fresh (default window 15 min). Dead/stale sessions do not count
+  toward blast radius.
+- **Leases / locks** — control-plane leases joined with work items and their
+  freshness (`lease_lifecycle.classify_lease_freshness`). Only `active` (live
+  owner) leases are *disruptive*; expired / released / dead-process leases never
+  withhold a restart.
+- **Issue / PR work** — the issues and PRs behind disruptive leases.
+- **Mutations / critical sections** — a live lease carrying an author worktree
+  or a mutating phase (`implementing`, `publishing`, `merging`, …) is a
+  critical section a restart must not sever.
+- **Terminal (merge) lock** — an active terminal lock always makes a restart
+  unsafe.
+- **Prior recovery attempts** — narrower recovery already tried (e.g. sanctioned
+  client reconnects) is echoed so the operator sees the escalation history.
+
+## Verdict
+
+Exactly three verdicts, matching the acceptance criteria:
+
+| Verdict | `allow_restart` | Meaning |
+|---------|-----------------|---------|
+| `safe` | `true` | No other live sessions, no live leases, no terminal lock. |
+| `unsafe` | `false` | Live work would be disrupted and no operator override is present — **or** the inventory could not be completed (fail closed). |
+| `override` | `true` | Live work present, but an operator override accepts the blast radius. |
+
+`override_would_allow` tells the console whether an override path exists for the
+current state. `blast_radius` is a `none` / `low` / `medium` / `high` severity
+band derived from the affected session and work counts.
+
+### Fail closed
+
+If the control-plane inventory cannot be completed (DB unavailable, a listing
+failed), `inventory_complete` is `false` and the verdict is `unsafe` / deny. An
+incomplete evaluation must never green-light a restart.
+
+### Operator override authority
+
+Override authority is read from the environment variable
+`GITEA_OPERATOR_RESTART_OVERRIDE_AUTHORIZATION` and **never** from a tool
+argument. A worker session cannot set an environment variable on an
+already-running daemon, so override cannot be self-asserted (same pattern as the
+#630 daemon-maintenance authorization). The `request_override` tool argument only
+expresses caller intent; it takes effect solely when the environment
+authorization is present.
+
+## The tool
+
+```text
+gitea_request_mcp_restart(remote, host, org, repo,
+                          dry_run=True, request_override=False,
+                          session_id=None, limit=200)
+```
+
+Read-only, dry-run, and it **never restarts anything**. `apply_supported` is
+always `false`; passing `dry_run=False` performs no restart and reports that
+apply is gated by a drain proof (a separate child).
+
+## Audit
+
+Every evaluation carries an `audit_record` (event, coordinator version, verdict,
+allow decision, blast radius, counts, timestamp) so restart decisions are
+auditable. No secrets flow through the coordinator — session ids, pids, and
+profiles are operational metadata only.
+
+A representative dry-run report is in
+[`mcp-restart-impact-sample.json`](./mcp-restart-impact-sample.json).

docs/mcp-restart-impact-sample.json

@@ -0,0 +1,148 @@
+{
+  "coordinator_version": "1.0.0-issue-658",
+  "evaluated_at": "2026-07-24T06:00:00+00:00",
+  "dry_run": true,
+  "restart_performed": false,
+  "inventory_complete": true,
+  "incomplete_reasons": [],
+  "verdict": "unsafe",
+  "allow_restart": false,
+  "override_would_allow": true,
+  "operator_override": false,
+  "blast_radius": "high",
+  "reasons": [
+    "live work would be disrupted; restart denied without operator override",
+    "1 critical section(s) in flight (active lease with a live owner)"
+  ],
+  "affected_sessions": [
+    {
+      "session_id": "prgs-author-30988-d6f43c25",
+      "role": "author",
+      "profile": "prgs-author",
+      "pid": 1,
+      "status": "active",
+      "alive": true,
+      "heartbeat_stale": false,
+      "is_requester": false,
+      "live": true
+    },
+    {
+      "session_id": "prgs-reviewer-4157-0ce9",
+      "role": "reviewer",
+      "profile": "prgs-reviewer",
+      "pid": 1,
+      "status": "active",
+      "alive": true,
+      "heartbeat_stale": false,
+      "is_requester": true,
+      "live": true
+    }
+  ],
+  "affected_leases": [
+    {
+      "lease_id": "lease-abc",
+      "session_id": "prgs-author-30988-d6f43c25",
+      "role": "author",
+      "phase": "implementing",
+      "freshness": "active",
+      "work_kind": "issue",
+      "work_number": 658,
+      "worktree_path": "/repo/branches/feat-issue-658",
+      "disruptive": true,
+      "is_mutation": true,
+      "is_critical_section": true
+    },
+    {
+      "lease_id": "lease-dead",
+      "session_id": "prgs-author-91485",
+      "role": "author",
+      "phase": "allocated",
+      "freshness": "stale_dead_process",
+      "work_kind": "issue",
+      "work_number": 651,
+      "worktree_path": null,
+      "disruptive": false,
+      "is_mutation": false,
+      "is_critical_section": false
+    }
+  ],
+  "critical_sections": [
+    {
+      "lease_id": "lease-abc",
+      "session_id": "prgs-author-30988-d6f43c25",
+      "role": "author",
+      "phase": "implementing",
+      "freshness": "active",
+      "work_kind": "issue",
+      "work_number": 658,
+      "worktree_path": "/repo/branches/feat-issue-658",
+      "disruptive": true,
+      "is_mutation": true,
+      "is_critical_section": true
+    }
+  ],
+  "affected_issues": [
+    658
+  ],
+  "affected_prs": [],
+  "mutations": [
+    {
+      "lease_id": "lease-abc",
+      "session_id": "prgs-author-30988-d6f43c25",
+      "role": "author",
+      "phase": "implementing",
+      "freshness": "active",
+      "work_kind": "issue",
+      "work_number": 658,
+      "worktree_path": "/repo/branches/feat-issue-658",
+      "disruptive": true,
+      "is_mutation": true,
+      "is_critical_section": true
+    }
+  ],
+  "terminal_lock": null,
+  "ack_state": {
+    "prgs-author-30988-d6f43c25": "pending"
+  },
+  "prior_recovery_attempts": [
+    {
+      "kind": "client_reconnect",
+      "at": "2026-07-24T06:00:00+00:00",
+      "outcome": "insufficient"
+    }
+  ],
+  "counts": {
+    "sessions_total": 2,
+    "sessions_live_other": 1,
+    "leases_total": 2,
+    "leases_disruptive": 1,
+    "critical_sections": 1,
+    "mutations": 1,
+    "affected_issues": 1,
+    "affected_prs": 0,
+    "prior_recovery_attempts": 1
+  },
+  "audit_record": {
+    "event": "restart_impact_evaluated",
+    "coordinator_version": "1.0.0-issue-658",
+    "evaluated_at": "2026-07-24T06:00:00+00:00",
+    "dry_run": true,
+    "operator_override": false,
+    "requesting_session_id": "prgs-reviewer-4157-0ce9",
+    "inventory_complete": true,
+    "verdict": "unsafe",
+    "allow_restart": false,
+    "blast_radius": "high",
+    "counts": {
+      "sessions_total": 2,
+      "sessions_live_other": 1,
+      "leases_total": 2,
+      "leases_disruptive": 1,
+      "critical_sections": 1,
+      "mutations": 1,
+      "affected_issues": 1,
+      "affected_prs": 0,
+      "prior_recovery_attempts": 1
+    }
+  }
+}

docs/mcp-tool-inventory.md

@@ -135,6 +135,7 @@ that gates each call, not which tools exist.
 - `gitea_release_merger_pr_lease`
 - `gitea_release_reviewer_pr_lease`
 - `gitea_release_workflow_lease`
+- `gitea_request_mcp_restart`
 - `gitea_resolve_task_capability`
 - `gitea_resume_review_draft`
 - `gitea_review_pr`

gitea_mcp_server.py

@@ -2040,6 +2040,7 @@ import dependency_graph  # noqa: E402  # #784 durable dependency edges
 import control_plane_db  # noqa: E402
 import lease_lifecycle  # noqa: E402
 import workflow_dashboard  # noqa: E402  # #605 live queue/lease dashboard
+import restart_coordinator  # noqa: E402  # #658 MCP restart coordinator/impact
 import incident_bridge  # noqa: E402
 import sentry_observability  # noqa: E402  (#606 optional Sentry observability)
 import sentry_incident_bridge  # noqa: E402  (#607 Sentry→Gitea incident bridge)
@@ -21489,6 +21490,156 @@ def gitea_workflow_dashboard(
     return payload
 
 
+@mcp.tool()
+def gitea_request_mcp_restart(
+    remote: str = "dadeschools",
+    host: str | None = None,
+    org: str | None = None,
+    repo: str | None = None,
+    dry_run: bool = True,
+    request_override: bool = False,
+    session_id: str | None = None,
+    limit: int = 200,
+) -> dict:
+    """Evaluate a proposed MCP restart and return an impact preview (#658).
+
+    Central restart coordinator: gathers live control-plane state (sessions,
+    leases/locks, in-flight issue/PR work, mutations, worktrees) and returns a
+    blast-radius impact report with a ``safe`` / ``unsafe`` / ``override``
+    verdict, so the console (#642/#652) and operators can see what a restart
+    would disrupt *before* any concurrent LLM work is destroyed.
+
+    This tool is **dry-run and never restarts anything.** The mutative apply
+    path is a separate child gated by a drain proof (non-goal here); calling
+    with ``dry_run=False`` still performs no restart and reports that apply is
+    not yet available.
+
+    Operator override authority is read from the process environment
+    (``GITEA_OPERATOR_RESTART_OVERRIDE_AUTHORIZATION``), never self-asserted by
+    the requesting session: ``request_override`` only expresses caller intent
+    and takes effect solely when that environment authorization is present.
+
+    Fails closed: if the control-plane inventory cannot be completed, the
+    verdict is ``unsafe`` / deny (an incomplete evaluation must never green-light
+    a restart).
+    """
+    read_block = _profile_operation_gate("gitea.read")
+    if read_block:
+        return {
+            "success": False,
+            "read_only": True,
+            "dry_run": True,
+            "restart_performed": False,
+            "reasons": read_block,
+            "permission_report": _permission_block_report("gitea.read"),
+        }
+
+    try:
+        h, o, r = _resolve(remote, host, org, repo)
+    except ValueError as exc:
+        return {
+            "success": False,
+            "read_only": True,
+            "dry_run": True,
+            "restart_performed": False,
+            "reasons": [str(exc)],
+        }
+
+    inventory_complete = True
+    incomplete_reasons: list[str] = []
+    sessions: list[dict] = []
+    leases: list[dict] = []
+    terminal_lock: dict | None = None
+
+    db, db_errs = _control_plane_db_or_error()
+    if db is None:
+        inventory_complete = False
+        incomplete_reasons.extend(
+            db_errs or ["control-plane DB unavailable; cannot evaluate restart"]
+        )
+    else:
+        try:
+            sessions = db.list_sessions(statuses=("active",), limit=max(1, int(limit)))
+        except Exception as exc:  # noqa: BLE001
+            inventory_complete = False
+            incomplete_reasons.append(
+                f"session inventory failed: {_redact(str(exc))}"
+            )
+        try:
+            lease_result = lease_lifecycle.list_active_leases(
+                db,
+                remote=remote if remote in REMOTES else remote,
+                org=o,
+                repo=r,
+                role=None,
+                include_non_active=False,
+                limit=max(1, int(limit)),
+            )
+            leases = list(lease_result.get("leases") or [])
+        except Exception as exc:  # noqa: BLE001
+            inventory_complete = False
+            incomplete_reasons.append(
+                f"lease inventory failed: {_redact(str(exc))}"
+            )
+        try:
+            terminal = db.get_active_terminal_lock(
+                remote=remote if remote in REMOTES else remote,
+                org=o,
+                repo=r,
+            )
+            if terminal:
+                terminal_lock = dict(terminal)
+        except Exception as exc:  # noqa: BLE001
+            inventory_complete = False
+            incomplete_reasons.append(
+                f"terminal lock lookup failed: {_redact(str(exc))}"
+            )
+
+    profile = get_profile()
+    profile_name = (profile.get("profile_name") or "").strip() or "session"
+    sid = (session_id or "").strip() or f"{profile_name}-{os.getpid()}"
+
+    # Override authority is read from the environment only — a worker session
+    # cannot set an env var for an already-running daemon, so it cannot be
+    # self-asserted the way a tool argument could (#630/#710 F1 pattern).
+    operator_authorized = bool(
+        (os.environ.get("GITEA_OPERATOR_RESTART_OVERRIDE_AUTHORIZATION") or "").strip()
+    )
+    operator_override = bool(request_override and operator_authorized)
+
+    inventory = {
+        "sessions": sessions,
+        "leases": leases,
+        "terminal_lock": terminal_lock,
+        "inventory_complete": inventory_complete,
+        "incomplete_reasons": incomplete_reasons,
+    }
+
+    report = restart_coordinator.evaluate_restart_impact(
+        inventory,
+        operator_override=operator_override,
+        requesting_session_id=sid,
+        dry_run=True,  # coordinator is always analysis-only (#658)
+    )
+
+    payload = report.as_dict()
+    payload["success"] = True
+    payload["read_only"] = True
+    payload["remote"] = remote
+    payload["org"] = o
+    payload["repo"] = r
+    payload["requesting_session_id"] = sid
+    payload["operator_override_requested"] = bool(request_override)
+    payload["operator_override_authorized"] = operator_authorized
+    payload["apply_supported"] = False
+    if not dry_run:
+        payload["reasons"] = list(payload.get("reasons") or []) + [
+            "apply requested but not supported: sanctioned restart apply is "
+            "gated by a drain proof (separate child); no restart performed (#658)"
+        ]
+    return payload
+
+
 @mcp.tool()
 def gitea_inspect_workflow_lease(
     lease_id: str,

restart_coordinator.py

@@ -0,0 +1,451 @@
+"""MCP restart coordinator and impact analysis (#658).
+
+Before any sanctioned MCP restart, a central coordinator must evaluate the
+live control-plane state — active sessions, leases/locks, in-flight issue/PR
+work, mutations, worktrees, and recovery history — and produce an *impact
+preview* so operators (and the web console, #642/#652) can see the blast
+radius **before** concurrent LLM work is disrupted.
+
+Design rules (mirrors the read-only posture of ``workflow_dashboard`` /
+``lease_lifecycle``):
+
+* **Pure classification.** :func:`evaluate_restart_impact` takes an already
+  gathered inventory and returns a structured report. It never touches the
+  network, the filesystem, or a live process, so multi-session fixtures can
+  drive every branch in unit tests. The coordinator *never restarts anything*;
+  a mutative apply path is a later child gated by a drain proof (non-goal here).
+* **Fail closed.** If the inventory is not explicitly complete, the verdict is
+  ``unsafe`` / deny — an incomplete evaluation must never green-light a restart.
+* **No secrets.** Session ids, pids, and profiles are operational metadata, not
+  credentials; nothing secret flows through this module.
+
+The single sanctioned entry point post-#657 is the MCP tool
+``gitea_request_mcp_restart`` (dry-run by default), which gathers the inventory
+from the #613 control-plane DB and calls :func:`evaluate_restart_impact`.
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from datetime import datetime, timezone
+from typing import Any, Mapping, Sequence
+
+import lease_lifecycle
+
+COORDINATOR_VERSION = "1.0.0-issue-658"
+
+# Restart verdicts. Exactly the three the acceptance criteria name.
+VERDICT_SAFE = "safe"
+VERDICT_UNSAFE = "unsafe"
+VERDICT_OVERRIDE = "override"
+
+# Blast-radius severity bands.
+BLAST_NONE = "none"
+BLAST_LOW = "low"
+BLAST_MEDIUM = "medium"
+BLAST_HIGH = "high"
+
+# A live lease with a live owner process is treated as active in-flight work.
+LEASE_FRESHNESS_LIVE = "active"
+
+# Default staleness window for a session heartbeat (seconds). A session whose
+# last heartbeat is older than this is not counted as live even if its row is
+# still marked ``active`` — it is assumed dead/detached.
+DEFAULT_SESSION_HEARTBEAT_STALE_SECONDS = 900
+
+
+def _utc_now() -> datetime:
+    return datetime.now(timezone.utc)
+
+
+def _parse_ts(value: str | None) -> datetime | None:
+    return lease_lifecycle._parse_ts(value)
+
+
+@dataclass(frozen=True)
+class SessionImpact:
+    """One MCP session a restart would terminate."""
+
+    session_id: str
+    role: str | None
+    profile: str | None
+    pid: int | None
+    status: str | None
+    alive: bool | None
+    heartbeat_stale: bool
+    is_requester: bool
+    live: bool
+
+    def as_dict(self) -> dict[str, Any]:
+        return {
+            "session_id": self.session_id,
+            "role": self.role,
+            "profile": self.profile,
+            "pid": self.pid,
+            "status": self.status,
+            "alive": self.alive,
+            "heartbeat_stale": self.heartbeat_stale,
+            "is_requester": self.is_requester,
+            "live": self.live,
+        }
+
+
+@dataclass(frozen=True)
+class LeaseImpact:
+    """One control-plane lease a restart would disrupt."""
+
+    lease_id: str | None
+    session_id: str | None
+    role: str | None
+    phase: str | None
+    freshness: str | None
+    work_kind: str | None
+    work_number: int | None
+    worktree_path: str | None
+    disruptive: bool
+    is_mutation: bool
+    is_critical_section: bool
+
+    def as_dict(self) -> dict[str, Any]:
+        return {
+            "lease_id": self.lease_id,
+            "session_id": self.session_id,
+            "role": self.role,
+            "phase": self.phase,
+            "freshness": self.freshness,
+            "work_kind": self.work_kind,
+            "work_number": self.work_number,
+            "worktree_path": self.worktree_path,
+            "disruptive": self.disruptive,
+            "is_mutation": self.is_mutation,
+            "is_critical_section": self.is_critical_section,
+        }
+
+
+@dataclass(frozen=True)
+class RestartImpactReport:
+    """Impact preview DTO returned to the console / operator (#642/#652)."""
+
+    coordinator_version: str
+    evaluated_at: str
+    dry_run: bool
+    restart_performed: bool
+    inventory_complete: bool
+    verdict: str
+    allow_restart: bool
+    override_would_allow: bool
+    operator_override: bool
+    blast_radius: str
+    reasons: list[str]
+    affected_sessions: list[SessionImpact]
+    affected_leases: list[LeaseImpact]
+    critical_sections: list[LeaseImpact]
+    affected_issues: list[int]
+    affected_prs: list[int]
+    mutations: list[LeaseImpact]
+    terminal_lock: dict[str, Any] | None
+    ack_state: dict[str, str]
+    prior_recovery_attempts: list[dict[str, Any]]
+    counts: dict[str, int]
+    audit_record: dict[str, Any]
+    incomplete_reasons: list[str] = field(default_factory=list)
+
+    def as_dict(self) -> dict[str, Any]:
+        return {
+            "coordinator_version": self.coordinator_version,
+            "evaluated_at": self.evaluated_at,
+            "dry_run": self.dry_run,
+            "restart_performed": self.restart_performed,
+            "inventory_complete": self.inventory_complete,
+            "incomplete_reasons": list(self.incomplete_reasons),
+            "verdict": self.verdict,
+            "allow_restart": self.allow_restart,
+            "override_would_allow": self.override_would_allow,
+            "operator_override": self.operator_override,
+            "blast_radius": self.blast_radius,
+            "reasons": list(self.reasons),
+            "affected_sessions": [s.as_dict() for s in self.affected_sessions],
+            "affected_leases": [l.as_dict() for l in self.affected_leases],
+            "critical_sections": [l.as_dict() for l in self.critical_sections],
+            "affected_issues": list(self.affected_issues),
+            "affected_prs": list(self.affected_prs),
+            "mutations": [l.as_dict() for l in self.mutations],
+            "terminal_lock": self.terminal_lock,
+            "ack_state": dict(self.ack_state),
+            "prior_recovery_attempts": list(self.prior_recovery_attempts),
+            "counts": dict(self.counts),
+            "audit_record": dict(self.audit_record),
+        }
+
+
+def _classify_session(
+    row: Mapping[str, Any],
+    *,
+    now: datetime,
+    requesting_session_id: str | None,
+    heartbeat_stale_seconds: int,
+) -> SessionImpact:
+    session_id = str(row.get("session_id") or "")
+    pid = row.get("pid")
+    status = (row.get("status") or "").strip().lower() or None
+    alive = lease_lifecycle.is_process_alive(pid) if pid is not None else None
+    hb = _parse_ts(row.get("last_heartbeat_at"))
+    heartbeat_stale = bool(
+        hb is not None and (now - hb).total_seconds() > heartbeat_stale_seconds
+    )
+    live = bool(status == "active" and alive is not False and not heartbeat_stale)
+    return SessionImpact(
+        session_id=session_id,
+        role=row.get("role"),
+        profile=row.get("profile"),
+        pid=pid,
+        status=status,
+        alive=alive,
+        heartbeat_stale=heartbeat_stale,
+        is_requester=bool(
+            requesting_session_id and session_id == requesting_session_id
+        ),
+        live=live,
+    )
+
+
+# Lease phases that represent an active mutation in flight (as opposed to a
+# mere allocation/claim with no work committed yet). An active lease in any of
+# these phases is a critical section a restart must not sever.
+_MUTATING_PHASES = frozenset(
+    {
+        "implementing",
+        "publishing",
+        "pushing",
+        "committing",
+        "reviewing",
+        "merging",
+        "reconciling",
+        "conflict_fix",
+    }
+)
+
+
+def _classify_lease(row: Mapping[str, Any]) -> LeaseImpact:
+    freshness_obj = row.get("freshness")
+    if isinstance(freshness_obj, Mapping):
+        freshness = str(freshness_obj.get("freshness") or "").strip().lower() or None
+    else:
+        freshness = str(freshness_obj or "").strip().lower() or None
+    phase = (row.get("phase") or "").strip().lower() or None
+    worktree = row.get("worktree_path")
+    disruptive = freshness == LEASE_FRESHNESS_LIVE
+    # A live lease is a mutation-in-flight if it carries an author worktree or
+    # its phase names a mutating step. All disruptive leases are critical
+    # sections a restart would sever regardless.
+    is_mutation = bool(
+        disruptive and (bool(worktree) or (phase in _MUTATING_PHASES))
+    )
+    number = row.get("work_number")
+    try:
+        number = int(number) if number is not None else None
+    except (TypeError, ValueError):
+        number = None
+    return LeaseImpact(
+        lease_id=row.get("lease_id"),
+        session_id=row.get("session_id"),
+        role=row.get("role"),
+        phase=phase,
+        freshness=freshness,
+        work_kind=(str(row.get("work_kind") or "").strip().lower() or None),
+        work_number=number,
+        worktree_path=worktree,
+        disruptive=disruptive,
+        is_mutation=is_mutation,
+        is_critical_section=disruptive,
+    )
+
+
+def _blast_radius(*, session_count: int, work_count: int, mutation_count: int) -> str:
+    if mutation_count > 0 or work_count >= 3 or session_count >= 3:
+        return BLAST_HIGH
+    if work_count > 0 or session_count == 2:
+        return BLAST_MEDIUM
+    if session_count == 1:
+        return BLAST_LOW
+    return BLAST_NONE
+
+
+def evaluate_restart_impact(
+    inventory: Mapping[str, Any],
+    *,
+    now: datetime | None = None,
+    operator_override: bool = False,
+    requesting_session_id: str | None = None,
+    dry_run: bool = True,
+    session_heartbeat_stale_seconds: int = DEFAULT_SESSION_HEARTBEAT_STALE_SECONDS,
+) -> RestartImpactReport:
+    """Evaluate a proposed MCP restart and return an impact preview.
+
+    ``inventory`` is a mapping with:
+
+    * ``sessions`` — session rows (session_id, role, profile, pid, status,
+      last_heartbeat_at).
+    * ``leases`` — control-plane lease rows, each ideally carrying an enriched
+      ``freshness`` dict (as :func:`lease_lifecycle.list_active_leases` returns);
+      a bare string freshness is also accepted.
+    * ``terminal_lock`` — the active terminal (merge) lock, if any.
+    * ``prior_recovery_attempts`` — narrower recovery attempts already tried
+      (e.g. sanctioned reconnects) so the operator sees escalation history.
+    * ``inventory_complete`` — bool. **Must** be explicitly True; a missing or
+      falsy value forces a deny (fail closed).
+    * ``incomplete_reasons`` — optional reasons the inventory is incomplete.
+
+    The coordinator never restarts anything: ``restart_performed`` is always
+    False and the mutative apply path is a later drain-gated child.
+    """
+    moment = now or _utc_now()
+    reasons: list[str] = []
+
+    inventory_complete = bool(inventory.get("inventory_complete", False))
+    incomplete_reasons = [str(r) for r in (inventory.get("incomplete_reasons") or [])]
+
+    sessions_raw: Sequence[Mapping[str, Any]] = inventory.get("sessions") or []
+    leases_raw: Sequence[Mapping[str, Any]] = inventory.get("leases") or []
+    terminal_lock = inventory.get("terminal_lock") or None
+    prior_recovery_attempts = [
+        dict(a) for a in (inventory.get("prior_recovery_attempts") or [])
+    ]
+
+    session_impacts = [
+        _classify_session(
+            s,
+            now=moment,
+            requesting_session_id=requesting_session_id,
+            heartbeat_stale_seconds=session_heartbeat_stale_seconds,
+        )
+        for s in sessions_raw
+    ]
+    lease_impacts = [_classify_lease(l) for l in leases_raw]
+
+    # Only *other* live sessions and live leases constitute blast radius: a
+    # restart that would kill only the requesting session with no other work in
+    # flight is safe.
+    other_live_sessions = [
+        s for s in session_impacts if s.live and not s.is_requester
+    ]
+    disruptive_leases = [l for l in lease_impacts if l.disruptive]
+    critical_sections = [l for l in lease_impacts if l.is_critical_section]
+    mutations = [l for l in lease_impacts if l.is_mutation]
+
+    affected_issues = sorted(
+        {
+            l.work_number
+            for l in disruptive_leases
+            if l.work_kind == "issue" and l.work_number is not None
+        }
+    )
+    affected_prs = sorted(
+        {
+            l.work_number
+            for l in disruptive_leases
+            if l.work_kind == "pr" and l.work_number is not None
+        }
+    )
+
+    disruptive = bool(disruptive_leases or other_live_sessions or terminal_lock)
+
+    if not inventory_complete:
+        verdict = VERDICT_UNSAFE
+        allow_restart = False
+        reasons.append(
+            "inventory incomplete: restart evaluation cannot confirm blast "
+            "radius — deny (fail closed, #658)"
+        )
+        reasons.extend(incomplete_reasons)
+    elif not disruptive:
+        verdict = VERDICT_SAFE
+        allow_restart = True
+        reasons.append("no other live sessions, live leases, or terminal lock")
+    elif operator_override:
+        verdict = VERDICT_OVERRIDE
+        allow_restart = True
+        reasons.append(
+            "live work present; operator override accepts the blast radius"
+        )
+    else:
+        verdict = VERDICT_UNSAFE
+        allow_restart = False
+        reasons.append(
+            "live work would be disrupted; restart denied without operator "
+            "override"
+        )
+
+    if critical_sections and inventory_complete:
+        reasons.append(
+            f"{len(critical_sections)} critical section(s) in flight "
+            "(active lease with a live owner)"
+        )
+    if terminal_lock:
+        reasons.append("active terminal (merge) lock present")
+
+    override_would_allow = bool(inventory_complete and disruptive)
+
+    blast_radius = _blast_radius(
+        session_count=len(other_live_sessions),
+        work_count=len(affected_issues) + len(affected_prs),
+        mutation_count=len(mutations),
+    )
+
+    # Acknowledgement is a later child (drain protocol); expose per-session
+    # placeholders so the console can render the ack column now.
+    ack_state = {s.session_id: "pending" for s in other_live_sessions}
+
+    counts = {
+        "sessions_total": len(session_impacts),
+        "sessions_live_other": len(other_live_sessions),
+        "leases_total": len(lease_impacts),
+        "leases_disruptive": len(disruptive_leases),
+        "critical_sections": len(critical_sections),
+        "mutations": len(mutations),
+        "affected_issues": len(affected_issues),
+        "affected_prs": len(affected_prs),
+        "prior_recovery_attempts": len(prior_recovery_attempts),
+    }
+
+    audit_record = {
+        "event": "restart_impact_evaluated",
+        "coordinator_version": COORDINATOR_VERSION,
+        "evaluated_at": moment.isoformat(),
+        "dry_run": dry_run,
+        "operator_override": bool(operator_override),
+        "requesting_session_id": requesting_session_id,
+        "inventory_complete": inventory_complete,
+        "verdict": verdict,
+        "allow_restart": allow_restart,
+        "blast_radius": blast_radius,
+        "counts": counts,
+    }
+
+    return RestartImpactReport(
+        coordinator_version=COORDINATOR_VERSION,
+        evaluated_at=moment.isoformat(),
+        dry_run=dry_run,
+        restart_performed=False,
+        inventory_complete=inventory_complete,
+        verdict=verdict,
+        allow_restart=allow_restart,
+        override_would_allow=override_would_allow,
+        operator_override=bool(operator_override),
+        blast_radius=blast_radius,
+        reasons=reasons,
+        affected_sessions=session_impacts,
+        affected_leases=lease_impacts,
+        critical_sections=critical_sections,
+        affected_issues=affected_issues,
+        affected_prs=affected_prs,
+        mutations=mutations,
+        terminal_lock=dict(terminal_lock)
+        if isinstance(terminal_lock, Mapping)
+        else terminal_lock,
+        ack_state=ack_state,
+        prior_recovery_attempts=prior_recovery_attempts,
+        counts=counts,
+        audit_record=audit_record,
+        incomplete_reasons=incomplete_reasons,
+    )

tests/test_restart_coordinator.py

@@ -0,0 +1,340 @@
+"""Tests for the MCP restart coordinator and impact analysis (#658).
+
+Multi-session fixtures exercise every verdict branch: safe, unsafe (live work),
+override, and the fail-closed deny on incomplete inventory. Also covers the
+critical-section deny path and the new ``ControlPlaneDB.list_sessions``.
+"""
+
+from __future__ import annotations
+
+import os
+import tempfile
+import unittest
+from datetime import datetime, timedelta, timezone
+
+import restart_coordinator as rc
+from control_plane_db import ControlPlaneDB
+
+
+NOW = datetime(2026, 7, 24, 6, 0, 0, tzinfo=timezone.utc)
+
+
+def _ts(dt: datetime) -> str:
+    return dt.isoformat()
+
+
+def _live_pid() -> int:
+    return os.getpid()
+
+
+def _dead_pid() -> int:
+    # A pid that is essentially never alive. os.kill(0) on it raises
+    # ProcessLookupError → is_process_alive False.
+    return 2_000_000_000
+
+
+def _session(session_id, *, pid, status="active", heartbeat=None, role="author"):
+    return {
+        "session_id": session_id,
+        "role": role,
+        "profile": "prgs-author",
+        "pid": pid,
+        "status": status,
+        "last_heartbeat_at": _ts(heartbeat or NOW),
+    }
+
+
+def _lease(
+    lease_id,
+    *,
+    session_id,
+    freshness,
+    kind="issue",
+    number=658,
+    phase="allocated",
+    worktree=None,
+    role="author",
+):
+    return {
+        "lease_id": lease_id,
+        "session_id": session_id,
+        "role": role,
+        "phase": phase,
+        "work_kind": kind,
+        "work_number": number,
+        "worktree_path": worktree,
+        "freshness": {"freshness": freshness},
+    }
+
+
+class EvaluateRestartImpactTest(unittest.TestCase):
+    def test_incomplete_inventory_denies_fail_closed(self) -> None:
+        report = rc.evaluate_restart_impact(
+            {"inventory_complete": False, "incomplete_reasons": ["db down"]},
+            now=NOW,
+        )
+        self.assertEqual(report.verdict, rc.VERDICT_UNSAFE)
+        self.assertFalse(report.allow_restart)
+        self.assertFalse(report.restart_performed)
+        self.assertIn("db down", report.incomplete_reasons)
+        self.assertTrue(
+            any("fail closed" in reasoning for reasoning in report.reasons)
+        )
+
+    def test_missing_completeness_flag_denies(self) -> None:
+        # No inventory_complete key at all → treated as incomplete.
+        report = rc.evaluate_restart_impact({}, now=NOW)
+        self.assertEqual(report.verdict, rc.VERDICT_UNSAFE)
+        self.assertFalse(report.allow_restart)
+
+    def test_no_other_work_is_safe(self) -> None:
+        report = rc.evaluate_restart_impact(
+            {
+                "inventory_complete": True,
+                "sessions": [_session("requester", pid=_live_pid())],
+                "leases": [],
+            },
+            now=NOW,
+            requesting_session_id="requester",
+        )
+        self.assertEqual(report.verdict, rc.VERDICT_SAFE)
+        self.assertTrue(report.allow_restart)
+        self.assertEqual(report.blast_radius, rc.BLAST_NONE)
+        self.assertEqual(report.affected_issues, [])
+
+    def test_dead_foreign_session_and_lease_are_not_disruptive(self) -> None:
+        report = rc.evaluate_restart_impact(
+            {
+                "inventory_complete": True,
+                "sessions": [
+                    _session("requester", pid=_live_pid()),
+                    _session("dead", pid=_dead_pid()),
+                ],
+                "leases": [
+                    _lease("l-dead", session_id="dead", freshness="stale_dead_process")
+                ],
+            },
+            now=NOW,
+            requesting_session_id="requester",
+        )
+        self.assertEqual(report.verdict, rc.VERDICT_SAFE)
+        self.assertTrue(report.allow_restart)
+        self.assertEqual(report.counts["leases_disruptive"], 0)
+        self.assertEqual(report.counts["sessions_live_other"], 0)
+
+    def test_live_foreign_lease_denies_without_override(self) -> None:
+        report = rc.evaluate_restart_impact(
+            {
+                "inventory_complete": True,
+                "sessions": [
+                    _session("requester", pid=_live_pid()),
+                    _session("worker", pid=_live_pid()),
+                ],
+                "leases": [
+                    _lease(
+                        "l1",
+                        session_id="worker",
+                        freshness="active",
+                        worktree="/tmp/wt-658",
+                        phase="implementing",
+                    )
+                ],
+            },
+            now=NOW,
+            requesting_session_id="requester",
+        )
+        self.assertEqual(report.verdict, rc.VERDICT_UNSAFE)
+        self.assertFalse(report.allow_restart)
+        # Critical section detected: active lease with a live owner.
+        self.assertEqual(len(report.critical_sections), 1)
+        self.assertEqual(report.affected_issues, [658])
+        self.assertEqual(report.counts["mutations"], 1)
+        self.assertTrue(report.override_would_allow)
+        self.assertEqual(report.blast_radius, rc.BLAST_HIGH)
+        # Placeholder ack state for the affected session.
+        self.assertEqual(report.ack_state.get("worker"), "pending")
+
+    def test_operator_override_allows_despite_live_work(self) -> None:
+        inv = {
+            "inventory_complete": True,
+            "sessions": [
+                _session("requester", pid=_live_pid()),
+                _session("worker", pid=_live_pid()),
+            ],
+            "leases": [_lease("l1", session_id="worker", freshness="active")],
+        }
+        report = rc.evaluate_restart_impact(
+            inv,
+            now=NOW,
+            requesting_session_id="requester",
+            operator_override=True,
+        )
+        self.assertEqual(report.verdict, rc.VERDICT_OVERRIDE)
+        self.assertTrue(report.allow_restart)
+        self.assertFalse(report.restart_performed)
+
+    def test_deny_when_critical_section_open(self) -> None:
+        # A single live author lease in a mutating phase is a critical section
+        # that must deny an un-overridden restart.
+        report = rc.evaluate_restart_impact(
+            {
+                "inventory_complete": True,
+                "sessions": [_session("worker", pid=_live_pid())],
+                "leases": [
+                    _lease(
+                        "l1",
+                        session_id="worker",
+                        freshness="active",
+                        phase="merging",
+                        kind="pr",
+                        number=900,
+                    )
+                ],
+            },
+            now=NOW,
+            requesting_session_id="requester",
+        )
+        self.assertEqual(report.verdict, rc.VERDICT_UNSAFE)
+        self.assertFalse(report.allow_restart)
+        self.assertEqual(report.affected_prs, [900])
+        self.assertEqual(len(report.critical_sections), 1)
+
+    def test_terminal_lock_makes_restart_unsafe(self) -> None:
+        report = rc.evaluate_restart_impact(
+            {
+                "inventory_complete": True,
+                "sessions": [_session("requester", pid=_live_pid())],
+                "leases": [],
+                "terminal_lock": {"terminal_pr": 812},
+            },
+            now=NOW,
+            requesting_session_id="requester",
+        )
+        self.assertEqual(report.verdict, rc.VERDICT_UNSAFE)
+        self.assertFalse(report.allow_restart)
+        self.assertIsNotNone(report.terminal_lock)
+        self.assertTrue(
+            any("terminal" in reasoning for reasoning in report.reasons)
+        )
+
+    def test_other_live_session_without_lease_is_disruptive(self) -> None:
+        report = rc.evaluate_restart_impact(
+            {
+                "inventory_complete": True,
+                "sessions": [
+                    _session("requester", pid=_live_pid()),
+                    _session("idle-but-live", pid=_live_pid()),
+                ],
+                "leases": [],
+            },
+            now=NOW,
+            requesting_session_id="requester",
+        )
+        self.assertEqual(report.verdict, rc.VERDICT_UNSAFE)
+        self.assertEqual(report.counts["sessions_live_other"], 1)
+
+    def test_stale_heartbeat_session_not_counted_live(self) -> None:
+        stale = NOW - timedelta(hours=2)
+        report = rc.evaluate_restart_impact(
+            {
+                "inventory_complete": True,
+                "sessions": [
+                    _session("requester", pid=_live_pid()),
+                    _session("stale", pid=_live_pid(), heartbeat=stale),
+                ],
+                "leases": [],
+            },
+            now=NOW,
+            requesting_session_id="requester",
+        )
+        self.assertEqual(report.verdict, rc.VERDICT_SAFE)
+        self.assertEqual(report.counts["sessions_live_other"], 0)
+
+    def test_prior_recovery_attempts_echoed(self) -> None:
+        report = rc.evaluate_restart_impact(
+            {
+                "inventory_complete": True,
+                "sessions": [_session("requester", pid=_live_pid())],
+                "leases": [],
+                "prior_recovery_attempts": [
+                    {"kind": "client_reconnect", "at": _ts(NOW)}
+                ],
+            },
+            now=NOW,
+            requesting_session_id="requester",
+        )
+        self.assertEqual(len(report.prior_recovery_attempts), 1)
+        self.assertEqual(report.counts["prior_recovery_attempts"], 1)
+
+    def test_bare_string_freshness_accepted(self) -> None:
+        lease = _lease("l1", session_id="worker", freshness="active")
+        lease["freshness"] = "active"  # bare string, not a dict
+        report = rc.evaluate_restart_impact(
+            {
+                "inventory_complete": True,
+                "sessions": [_session("worker", pid=_live_pid())],
+                "leases": [lease],
+            },
+            now=NOW,
+            requesting_session_id="requester",
+        )
+        self.assertEqual(report.counts["leases_disruptive"], 1)
+
+    def test_as_dict_is_serializable_dto(self) -> None:
+        import json
+
+        report = rc.evaluate_restart_impact(
+            {
+                "inventory_complete": True,
+                "sessions": [_session("requester", pid=_live_pid())],
+                "leases": [],
+            },
+            now=NOW,
+            requesting_session_id="requester",
+        )
+        payload = report.as_dict()
+        # Round-trips through JSON — safe for the console DTO.
+        encoded = json.dumps(payload)
+        decoded = json.loads(encoded)
+        self.assertEqual(decoded["verdict"], rc.VERDICT_SAFE)
+        self.assertIn("audit_record", decoded)
+        self.assertEqual(decoded["audit_record"]["event"], "restart_impact_evaluated")
+        self.assertFalse(decoded["restart_performed"])
+        self.assertIn("coordinator_version", decoded)
+
+
+class ListSessionsTest(unittest.TestCase):
+    def setUp(self) -> None:
+        self._tmp = tempfile.TemporaryDirectory()
+        self.db = ControlPlaneDB(os.path.join(self._tmp.name, "cp.sqlite3"))
+
+    def tearDown(self) -> None:
+        self._tmp.cleanup()
+
+    def test_list_sessions_filters_by_status(self) -> None:
+        self.db.upsert_session(session_id="a", role="author", pid=1, status="active")
+        self.db.upsert_session(session_id="b", role="author", pid=2, status="ended")
+        active = self.db.list_sessions(statuses=("active",))
+        ids = {row["session_id"] for row in active}
+        self.assertEqual(ids, {"a"})
+        every = self.db.list_sessions()
+        self.assertEqual({row["session_id"] for row in every}, {"a", "b"})
+
+    def test_list_sessions_feeds_coordinator(self) -> None:
+        self.db.upsert_session(
+            session_id="requester", role="author", pid=os.getpid(), status="active"
+        )
+        report = rc.evaluate_restart_impact(
+            {
+                "inventory_complete": True,
+                "sessions": self.db.list_sessions(statuses=("active",)),
+                "leases": [],
+            },
+            now=NOW,
+            requesting_session_id="requester",
+        )
+        self.assertEqual(report.counts["sessions_total"], 1)
+
+
+if __name__ == "__main__":  # pragma: no cover
+    unittest.main()