feat(launcher): trusted client-instance identity for project-scoped launches

Issue #985. The production launcher could mint one trusted
GITEA_MCP_CLIENT_INSTANCE per launch, but two gaps kept real launches on
untrusted legacy-pid-* identities:

1. build_application_mcp_servers required a profile for all five sanctioned
   namespaces, so a project-scoped configuration exposing only author,
   reviewer, and merger could not use it without inventing controller and
   reconciler profiles that must not exist.
2. The module produced configuration data but had no runnable entry point, so
   every real launch bypassed it entirely.

Changes:

- resolve_launch_namespaces() validates an explicit namespace subset as an
  allow-list; unknown, duplicate, and empty selections are refused rather than
  silently narrowing a launch. Omitting it preserves five-namespace behaviour.
- build_application_mcp_servers() accepts that subset, requires profiles only
  for the launched namespaces, starts only those workers, and reports
  excluded_namespaces / project_scoped.
- collect_instance_ids_from_mcp_servers() inspects the launch's own namespaces
  instead of an assumed five, and reports missing_servers, so a three-namespace
  launch can prove shared attribution without reading as two absent workers.
- Runnable entry point: python3 -m mcp_application_launcher mints one trusted
  identity, writes a per-launch 0600 mcpServers config, and execs the client.
  CLIENT_LAUNCH_SPECS is a data-driven registry so other supported clients use
  the same mint-once/propagate-to-all mechanism. --dry-run prints the plan.
- Provenance sealing now fails closed. The inst- format is public and
  reproducible, so format alone could previously let anyone who set one
  environment variable manufacture a trusted identity. Trust now additionally
  requires GITEA_MCP_INSTANCE_PROVENANCE=trusted_launcher, which only the
  launcher writes; a well-formed but unsealed value is classified
  unsealed_launcher and refused, while still being reported for diagnosis.

Deliberate behaviour change: tests/test_issue_978_instance_fleet_snapshot.py
test_client_hints_trusted_when_set previously asserted that a well-formed ID
alone was trusted. It now supplies the launcher seal, and a new companion test
asserts the unsealed case fails closed. This tightens the contract; no
assertion was weakened.

No static or persistent per-project instance IDs are introduced, duplicate
worker and cohort detection are untouched, and no fleet or mutation gate is
relaxed.

Tests: tests/test_issue_985_project_scoped_launcher.py, 47 passed, 3 subtests.
Full suite from a branches/ worktree: 28 failed, 6252 passed, 6 skipped against
a master baseline at 32ab8392 of 28 failed, 6204 passed, 6 skipped; the failing
sets are byte-identical, so zero regressions and zero masked failures.

Closes #985

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

anti_stomp_preflight.py

@@ -355,6 +355,10 @@ def assess_anti_stomp_preflight(
     root_head_sha: str | None = None,
     root_porcelain: str | None = None,
     remote_master_sha: str | None = None,
+    # #983: the tracking integration ref the SHA above came from, so root-checkout
+    # contamination names the ref actually compared instead of a hardcoded
+    # 'prgs/master'. None preserves the previous generic wording.
+    remote_master_ref: str | None = None,
     check_root_checkout: bool = True,
     check_worktree: bool = True,
     create_issue_bootstrap_assessment: dict[str, Any] | None = None,
@@ -553,6 +557,7 @@ def assess_anti_stomp_preflight(
             remote_master_sha=remote_master_sha,
             resolved_role=req_role or role,
             actual_role=role,
+            remote_master_ref=remote_master_ref,
         )
         checks["root_checkout"] = {
             "block": bool(root_assessment.get("block")),

canonical_repository_root.py

@@ -37,6 +37,34 @@ CANONICAL_ROOT_ENV = "GITEA_CANONICAL_REPOSITORY_ROOT"
 # Candidate git remote names probed when deriving repository identity.
 _IDENTITY_REMOTE_CANDIDATES = ("prgs", "origin", "dadeschools", "mdcps")
 
+# Fallback integration-branch names, probed only when the checkout declares no
+# configured upstream (#983). Mirrors the stable base branches recognised
+# elsewhere in the workflow (``stacked_pr_support``, ``root_checkout_guard``).
+# The fallback is deliberately *not* ordered-first-wins: when more than one of
+# these refs exists and the checkout records no upstream, the integration branch
+# is genuinely ambiguous and resolution fails closed instead of guessing.
+INTEGRATION_BRANCH_CANDIDATES: tuple[str, ...] = ("master", "main", "dev")
+
+# Sources for a *proven* base-ref derivation (#983 B1).
+#
+# ``refs/remotes/<remote>/HEAD`` is deliberately absent from this list. It is a
+# local symbolic-ref *cache* written once at clone time and refreshed only by an
+# explicit ``git remote set-head``; an ordinary fetch never updates it. When the
+# upstream default branch changes afterwards the cache keeps naming the old
+# branch, so trusting it derives the wrong integration branch for a checkout
+# that is sitting exactly on its tip. The cache is still read, but only as a
+# corroborating observation reported back to the caller — never as an authority,
+# and never as a tie-breaker between otherwise ambiguous candidates.
+BASE_REF_SOURCE_CONFIGURED_UPSTREAM = "configured_branch_upstream"
+BASE_REF_SOURCE_UNIQUE_CANDIDATE = "unique_integration_branch_ref"
+
+# Reason codes for base-ref derivation outcomes (#983), so callers and tests can
+# assert the refusal cause instead of string-matching prose.
+DENY_NO_IDENTITY_REMOTE = "no_identity_remote"
+DENY_AMBIGUOUS_REMOTE = "ambiguous_identity_remote"
+DENY_AMBIGUOUS_BASE_BRANCH = "ambiguous_integration_branch"
+DENY_NO_BASE_BRANCH = "no_integration_branch_ref"
+
 # Repository-authority modes (#973 B10). Exactly two values are supported.
 # ``mode`` selects how repository authority is established, so an unrecognised
 # value must never be normalised onto one of these: aliasing a trusted mode is
@@ -120,17 +148,15 @@ def resolve_repo_toplevel(path: str) -> str | None:
     return os.path.realpath(top) if top else None
 
 
-def repository_identity_slug(path: str, *, remote: str | None = None) -> str | None:
-    """``owner/repository`` derived from a git remote configured at *path*.
+def _identity_remote_candidates(path: str, remote: str | None) -> list[str]:
+    """Ordered remote names to probe for identity at *path*.
 
-    Tries the caller-named remote first, then a small set of known remote names,
-    then whatever remote the repository actually has. Returns None when no remote
-    URL is parseable (identity cannot be proven).
+    Names are used verbatim — never case-folded. Git config subsection names are
+    case-sensitive, so a repository whose remote is ``MDCPS`` is reached only by
+    the exact string ``MDCPS``; the lowercase entry in
+    :data:`_IDENTITY_REMOTE_CANDIDATES` simply does not resolve, and the exact
+    name arrives from ``git remote`` below (#983).
     """
-    text = (path or "").strip()
-    if not text:
-        return None
-
     ordered: list[str] = []
     for name in (remote, *_IDENTITY_REMOTE_CANDIDATES):
         clean = (name or "").strip()
@@ -139,7 +165,7 @@ def repository_identity_slug(path: str, *, remote: str | None = None) -> str | N
 
     try:
         listed = subprocess.run(
-            ["git", "-C", text, "remote"],
+            ["git", "-C", path, "remote"],
             capture_output=True,
             text=True,
             check=True,
@@ -149,11 +175,20 @@ def repository_identity_slug(path: str, *, remote: str | None = None) -> str | N
     for name in listed:
         if name and name not in ordered:
             ordered.append(name)
+    return ordered
 
-    for name in ordered:
+
+def _configured_remote_identities(path: str, remote: str | None) -> list[tuple[str, str]]:
+    """``(remote_name, owner/repository)`` for every probe name that resolves.
+
+    Remote names are returned exactly as configured so downstream tracking refs
+    (``refs/remotes/<remote>/<branch>``) address the real ref (#983).
+    """
+    found: list[tuple[str, str]] = []
+    for name in _identity_remote_candidates(path, remote):
         try:
             url = subprocess.run(
-                ["git", "-C", text, "remote", "get-url", name],
+                ["git", "-C", path, "remote", "get-url", name],
                 capture_output=True,
                 text=True,
                 check=True,
@@ -162,8 +197,415 @@ def repository_identity_slug(path: str, *, remote: str | None = None) -> str | N
             continue
         parsed = remote_repo_guard.parse_org_repo_from_remote_url(url)
         if parsed:
-            return f"{parsed[0]}/{parsed[1]}"
-    return None
+            found.append((name, f"{parsed[0]}/{parsed[1]}"))
+    return found
+
+
+def _configured_branch_upstream(path: str) -> tuple[str | None, str | None]:
+    """``(remote_name, branch)`` the checked-out branch is configured to track.
+
+    Reads ``branch.<current>.remote`` and ``branch.<current>.merge`` — the
+    checkout's own explicitly configured integration target, equivalent to
+    ``@{upstream}``. Unlike ``refs/remotes/<remote>/HEAD`` this is not a
+    clone-time cache: it is written when the branch is set up to track an
+    upstream and rewritten whenever that tracking changes, so it states what the
+    checkout actually integrates onto today (#983 B1).
+
+    Returns ``(None, None)`` on a detached HEAD or an untracked branch. Names are
+    returned verbatim; git remote names are case-sensitive.
+    """
+    text = (path or "").strip()
+    if not text:
+        return None, None
+
+    branch = _git_read(text, "symbolic-ref", "--quiet", "--short", "HEAD")
+    if not branch:
+        return None, None
+
+    remote = _git_read(text, "config", "--get", f"branch.{branch}.remote")
+    merge = _git_read(text, "config", "--get", f"branch.{branch}.merge")
+    if not remote or not merge:
+        return None, None
+
+    prefix = "refs/heads/"
+    upstream_branch = merge[len(prefix):].strip() if merge.startswith(prefix) else merge.strip()
+    if not upstream_branch:
+        return None, None
+    return remote, upstream_branch
+
+
+def _cached_remote_head_branch(path: str, remote: str) -> str | None:
+    """Branch named by the *cached* ``refs/remotes/<remote>/HEAD`` symref.
+
+    Read for observability only. This value is never authoritative (see
+    :data:`BASE_REF_SOURCE_CONFIGURED_UPSTREAM`); it is surfaced so an operator
+    can see that the local cache disagrees with the configured upstream, and so
+    a refusal can name the misleading signal explicitly.
+    """
+    prefix = f"refs/remotes/{remote}/"
+    symref = _git_read(path, "symbolic-ref", "--quiet", f"{prefix}HEAD")
+    if not symref or not symref.startswith(prefix):
+        return None
+    branch = symref[len(prefix):].strip()
+    return branch or None
+
+
+def assess_identity_remote(
+    path: str, *, explicit_remote: str | None = None
+) -> dict:
+    """Resolve the identity remote, failing closed when the target is ambiguous.
+
+    This is the ambiguity-aware counterpart to :func:`resolve_identity_remote`,
+    and the reason gating and reporting can no longer disagree (#983 B2).
+
+    *explicit_remote* is a **caller-supplied disambiguation** and nothing else.
+    It must come from an operator or an explicitly sanctioned repository
+    context; a value this module inferred while probing must never be handed
+    back in through it, because doing so re-labels an internal first-wins guess
+    as deliberate caller intent and silently suppresses the ambiguity gate.
+
+    Decision table:
+
+    * No remote yields a parseable identity -> :data:`DENY_NO_IDENTITY_REMOTE`.
+    * *explicit_remote* names one of the resolving remotes -> that remote is
+      authoritative, ``explicit`` True.
+    * Otherwise, when every resolving remote claims the **same** repository the
+      target is unambiguous and is accepted, ``explicit`` False. The remote
+      named by the checkout's configured upstream is preferred among equals so
+      the choice is deterministic rather than probe-order dependent.
+    * Otherwise distinct remotes claim different repositories and there is no
+      sanctioned disambiguation -> :data:`DENY_AMBIGUOUS_REMOTE`.
+
+    Returns a dict with ``remote``, ``slug``, ``identities``, ``ambiguous``,
+    ``explicit``, ``reason_code`` and ``reasons``. ``remote``/``slug`` are None
+    on any refusal, so a caller cannot report a repository the gate refuses.
+    """
+    text = (path or "").strip()
+    result: dict = {
+        "remote": None,
+        "slug": None,
+        "identities": [],
+        "ambiguous": False,
+        "explicit": False,
+        "reason_code": None,
+        "reasons": [],
+    }
+    if not text:
+        result["reason_code"] = DENY_NO_IDENTITY_REMOTE
+        result["reasons"].append(
+            "no repository path supplied for identity-remote resolution (fail closed)"
+        )
+        return result
+
+    named = (explicit_remote or "").strip() or None
+    identities = _configured_remote_identities(text, named)
+    result["identities"] = list(identities)
+    if not identities:
+        result["reason_code"] = DENY_NO_IDENTITY_REMOTE
+        result["reasons"].append(
+            f"no git remote at '{text}' yields a parseable repository identity "
+            "(fail closed)"
+        )
+        return result
+
+    if named:
+        for name, slug in identities:
+            if name == named:
+                result["remote"] = name
+                result["slug"] = slug
+                result["explicit"] = True
+                return result
+
+    distinct = {slug for _, slug in identities}
+    if len(distinct) > 1:
+        listed = ", ".join(f"{name} -> {slug}" for name, slug in identities)
+        result["ambiguous"] = True
+        result["reason_code"] = DENY_AMBIGUOUS_REMOTE
+        detail = (
+            f"explicitly named remote '{named}' does not resolve a repository identity "
+            "there, so it cannot disambiguate; "
+            if named
+            else ""
+        )
+        result["reasons"].append(
+            f"ambiguous repository identity at '{text}': {detail}remotes resolve to "
+            f"different repositories ({listed}); no single authoritative target can "
+            "be established (fail closed)"
+        )
+        return result
+
+    # One repository, possibly reachable through several remote names (a mirror).
+    # Prefer the remote the checkout is actually configured to track so the
+    # choice is deterministic instead of probe-order dependent.
+    upstream_remote, _ = _configured_branch_upstream(text)
+    chosen = identities[0]
+    if upstream_remote:
+        for entry in identities:
+            if entry[0] == upstream_remote:
+                chosen = entry
+                break
+    result["remote"], result["slug"] = chosen
+    return result
+
+
+def resolve_identity_remote(
+    path: str, *, remote: str | None = None
+) -> tuple[str | None, str | None]:
+    """``(remote_name, owner/repository)`` for the remote that proves identity.
+
+    The remote *name* is the piece historically thrown away by
+    :func:`repository_identity_slug`, even though resolving the slug already
+    required discovering it. Cross-repository base-ref derivation needs that
+    name to build ``refs/remotes/<remote>/<branch>``, so it is now returned
+    rather than discarded (#983). Returns ``(None, None)`` when no remote URL is
+    parseable (identity cannot be proven).
+
+    First-wins by design: this is the identity lookup behind
+    :func:`repository_identity_slug` and the #706/#973 canonical-root
+    validation, which compare an observed slug against an independently trusted
+    expected slug and therefore do not need an ambiguity verdict. Callers that
+    *derive* a target rather than validate one — the mutation guard and the
+    parity report — must use :func:`assess_identity_remote`, which fails closed
+    on ambiguity (#983 B2).
+    """
+    text = (path or "").strip()
+    if not text:
+        return None, None
+    for name, slug in _configured_remote_identities(text, remote):
+        return name, slug
+    return None, None
+
+
+def repository_identity_slug(path: str, *, remote: str | None = None) -> str | None:
+    """``owner/repository`` derived from a git remote configured at *path*.
+
+    Tries the caller-named remote first, then a small set of known remote names,
+    then whatever remote the repository actually has. Returns None when no remote
+    URL is parseable (identity cannot be proven).
+    """
+    return resolve_identity_remote(path, remote=remote)[1]
+
+
+def _base_ref_result(
+    *,
+    proven: bool,
+    reasons: list[str],
+    remote: str | None = None,
+    branch: str | None = None,
+    repository_slug: str | None = None,
+    source: str | None = None,
+    reason_code: str | None = None,
+    identity_explicit: bool = False,
+    configured_upstream_remote: str | None = None,
+    configured_upstream_branch: str | None = None,
+    cached_remote_head_branch: str | None = None,
+) -> dict:
+    """Build the base-ref derivation payload.
+
+    ``tracking_refs`` is the ordered probe tuple downstream guards hand to
+    ``git rev-parse``: the ``<remote>/<branch>`` shorthand first, then the fully
+    qualified ``refs/remotes/<remote>/<branch>``. It is empty whenever the
+    derivation is not ``proven``, so an unresolved target can never be probed
+    against some other repository's ref.
+
+    ``cached_remote_head_branch`` reports what the local
+    ``refs/remotes/<remote>/HEAD`` cache claims, and
+    ``cached_remote_head_conflicts`` whether that claim disagrees with the branch
+    actually derived. Both are observability only: the cache never decides the
+    outcome (#983 B1).
+    """
+    tracking_ref = f"refs/remotes/{remote}/{branch}" if proven and remote and branch else None
+    tracking_refs: tuple[str, ...] = (
+        (f"{remote}/{branch}", tracking_ref) if tracking_ref else ()
+    )
+    return {
+        "proven": proven,
+        "block": not proven,
+        "remote": remote,
+        "branch": branch,
+        "repository_slug": repository_slug,
+        "tracking_ref": tracking_ref,
+        "tracking_refs": tracking_refs,
+        "source": source,
+        "reason_code": reason_code,
+        "identity_explicit": identity_explicit,
+        "configured_upstream_remote": configured_upstream_remote,
+        "configured_upstream_branch": configured_upstream_branch,
+        "cached_remote_head_branch": cached_remote_head_branch,
+        "cached_remote_head_conflicts": bool(
+            cached_remote_head_branch and branch and cached_remote_head_branch != branch
+        ),
+        "reasons": list(reasons),
+    }
+
+
+def _git_read(path: str, *args: str) -> str | None:
+    """Run a read-only git command in *path*; ``None`` on any failure."""
+    try:
+        res = subprocess.run(
+            ["git", "-C", path, *args],
+            capture_output=True,
+            text=True,
+            check=False,
+        )
+    except Exception:
+        return None
+    if res.returncode != 0:
+        return None
+    return (res.stdout or "").strip() or None
+
+
+def _ref_exists(path: str, ref: str) -> bool:
+    """Whether *ref* resolves in the checkout at *path*."""
+    try:
+        res = subprocess.run(
+            ["git", "-C", path, "rev-parse", "--verify", "--quiet", ref],
+            capture_output=True,
+            text=True,
+            check=False,
+        )
+    except Exception:
+        return False
+    return res.returncode == 0 and bool((res.stdout or "").strip())
+
+
+def resolve_target_base_ref(path: str, *, explicit_remote: str | None = None) -> dict:
+    """Derive the authoritative integration base ref for the checkout at *path*.
+
+    This is the single resolved target shared by cross-repository mutation
+    gating and parity reporting, so the two can never disagree about which
+    commit a checkout is supposed to match (#983).
+
+    *explicit_remote* is a caller-supplied disambiguation only; see
+    :func:`assess_identity_remote` for why an internally inferred remote must
+    never be passed back in here (#983 B2).
+
+    Resolution order:
+
+    1. **Identity remote** — via :func:`assess_identity_remote`, with its exact
+       configured case preserved (``MDCPS`` stays ``MDCPS``). Ambiguous identity
+       fails closed here rather than resolving to whichever remote probed first.
+    2. **The checkout's configured upstream** — ``branch.<current>.remote`` plus
+       ``branch.<current>.merge``, accepted when it names the identity remote
+       and its remote-tracking ref actually exists. This is the checkout's own
+       declaration of what it integrates onto, and unlike the remote-HEAD cache
+       it is rewritten whenever that tracking changes.
+    3. **Fallback** — only when the checkout declares no usable upstream,
+       exactly one of :data:`INTEGRATION_BRANCH_CANDIDATES` present as a
+       remote-tracking ref.
+
+    ``refs/remotes/<remote>/HEAD`` is **not** a step. It is read for reporting
+    (``cached_remote_head_branch`` / ``cached_remote_head_conflicts``) and never
+    decides the branch, because it is a clone-time cache that an ordinary fetch
+    does not refresh: a checkout whose upstream default moved on still has the
+    old branch cached, and trusting it gates that checkout against a ref it does
+    not integrate onto (#983 B1).
+
+    Fails closed — ``proven`` False, empty ``tracking_refs``, and a
+    ``reason_code`` — when identity is unprovable, when distinct remotes claim
+    different repositories, when several candidate branches exist with no
+    configured upstream, or when no candidate exists at all. Nothing here
+    invents a branch, writes a ref, runs a fetch, or falls back to another
+    repository's base. Every ``proven`` result names a tracking ref that
+    resolves in this checkout.
+    """
+    text = (path or "").strip()
+    if not text:
+        return _base_ref_result(
+            proven=False,
+            reasons=["no repository path supplied for base-ref derivation (fail closed)"],
+            reason_code=DENY_NO_IDENTITY_REMOTE,
+        )
+
+    identity = assess_identity_remote(text, explicit_remote=explicit_remote)
+    remote_name, slug = identity["remote"], identity["slug"]
+    if not remote_name:
+        return _base_ref_result(
+            proven=False,
+            reasons=list(identity["reasons"]),
+            reason_code=identity["reason_code"],
+            identity_explicit=bool(identity["explicit"]),
+        )
+
+    prefix = f"refs/remotes/{remote_name}/"
+    cached = _cached_remote_head_branch(text, remote_name)
+    upstream_remote, upstream_branch = _configured_branch_upstream(text)
+    common = {
+        "repository_slug": slug,
+        "identity_explicit": bool(identity["explicit"]),
+        "configured_upstream_remote": upstream_remote,
+        "configured_upstream_branch": upstream_branch,
+        "cached_remote_head_branch": cached,
+    }
+
+    # 2. The checkout's configured upstream, when it belongs to the identity
+    #    remote and its tracking ref is actually present. Requiring the ref to
+    #    exist keeps 'proven' honest: a proven target is always resolvable.
+    if (
+        upstream_remote == remote_name
+        and upstream_branch
+        and _ref_exists(text, f"{prefix}{upstream_branch}")
+    ):
+        return _base_ref_result(
+            proven=True,
+            reasons=[],
+            remote=remote_name,
+            branch=upstream_branch,
+            source=BASE_REF_SOURCE_CONFIGURED_UPSTREAM,
+            **common,
+        )
+
+    # 3. Exactly one known integration branch present as a remote-tracking ref.
+    present = [
+        candidate
+        for candidate in INTEGRATION_BRANCH_CANDIDATES
+        if _ref_exists(text, f"{prefix}{candidate}")
+    ]
+    if len(present) == 1:
+        return _base_ref_result(
+            proven=True,
+            reasons=[],
+            remote=remote_name,
+            branch=present[0],
+            source=BASE_REF_SOURCE_UNIQUE_CANDIDATE,
+            **common,
+        )
+
+    # The cached remote HEAD is named in the refusal so the operator can see the
+    # signal that looks authoritative but is not, and is told the read-only fix.
+    cache_note = (
+        f" the cached '{prefix}HEAD' names '{cached}', but that cache is written at "
+        "clone time and is not refreshed by fetch, so it cannot break the tie;"
+        if cached
+        else ""
+    )
+    remedy = (
+        f" Configure the checkout's upstream (git branch --set-upstream-to={remote_name}/"
+        "<branch>) so the integration target is declared rather than guessed."
+    )
+    if len(present) > 1:
+        return _base_ref_result(
+            proven=False,
+            reasons=[
+                f"the checkout at '{text}' declares no upstream on remote "
+                f"'{remote_name}' and several integration branches exist "
+                f"({', '.join(present)});{cache_note} the integration base ref is "
+                f"ambiguous (fail closed).{remedy}"
+            ],
+            reason_code=DENY_AMBIGUOUS_BASE_BRANCH,
+            **common,
+        )
+    return _base_ref_result(
+        proven=False,
+        reasons=[
+            f"the checkout at '{text}' declares no upstream on remote '{remote_name}' "
+            f"and none of {'/'.join(INTEGRATION_BRANCH_CANDIDATES)} exists as a "
+            f"remote-tracking ref under '{prefix}';{cache_note} the integration base "
+            f"ref cannot be derived (fail closed; no fetch is performed here).{remedy}"
+        ],
+        reason_code=DENY_NO_BASE_BRANCH,
+        **common,
+    )
 
 
 def assess_canonical_repository_root(

docs/instance-fleet-identity.md

@@ -0,0 +1,286 @@
+# Instance-level fleet identity and health snapshots
+
+Issue **#978**. Companion primitives: **#948** (worker ownership), **#975**
+(heartbeat lifecycle), **#951** (restart receipts).
+
+## Why this exists
+
+The multi-instance fleet gate (#963) needs a *native* answer to:
+
+* which application launches are live;
+* which of the five namespace workers belong to each launch;
+* whether two Codex (or Claude, Grok, …) launches are distinct;
+* whether any live collision is real rather than a shared client type.
+
+Process tables, PID proximity, configuration files, and direct SQLite access
+are **not** production evidence. The sanctioned surface is the read-only MCP
+tool `gitea_snapshot_instance_fleet` on **controller** and **reconciler**
+namespaces.
+
+## Identity hierarchy
+
+| Identity | Scope | Who generates it | Lifetime |
+| --- | --- | --- | --- |
+| `client_type` | Application family (`codex`, `claude_code`, `gemini`, `grok`, …) | Launcher sets `GITEA_MCP_CLIENT` | Stable for the product |
+| `client_instance_id` | One running application launch | **Trusted launcher**, once per launch, as `GITEA_MCP_CLIENT_INSTANCE` | Fresh launch → new ID; reconnect of same launch → same ID; full app restart → new ID |
+| `fleet_run_id` | Operator-approved enrollment / canary cohort | Operator / controller sets `GITEA_MCP_FLEET_RUN_ID` | Duration of the approved rollout |
+| `namespace` | `author` \| `reviewer` \| `merger` \| `controller` \| `reconciler` | Profile / MCP server binding | Process lifetime |
+| `worker_id` / `worker_identity` | One namespace worker process | Runtime registry at first registration | New on worker restart; not reused |
+| `session_id` | Client session ownership | Launcher `GITEA_MCP_CLIENT_SESSION` or runtime | Session lifetime |
+| `generation_id` | One daemon launch | Runtime at process boot | New on process restart |
+| `process_identity` / PID | OS process | Runtime | Process lifetime |
+
+### Rules
+
+1. Multiple active instances **may** share the same `client_type`.
+2. Every application launch receives a **distinct** `client_instance_id`.
+3. All namespace workers of one launch report the **same**
+   `client_instance_id`. For a whole-fleet launch that is five workers; for a
+   project-scoped launch it is exactly the namespaces that launch started
+   (#985).
+4. Each namespace worker has a **distinct** `worker_identity`, process
+   identity, generation, and PID.
+5. Instance identity is **never** inferred from PID proximity, timestamps, or
+   client type alone.
+6. Live reuse of one `client_instance_id` with conflicting workers fails closed.
+7. Sharing only a profile or `client_type` is **not** a duplicate.
+
+This deliberately **replaces** any permanent `exactly_one_per_profile` fleet
+model (#949 assumption) as the operating rule for multi-instance fleets.
+
+## Project-scoped launches and the runnable entry point (#985)
+
+Not every project exposes all five namespaces. A project-scoped configuration
+such as Weekly Briefings intentionally has **author, reviewer, and merger
+only**, and has no controller or reconciler profile to supply. Requiring all
+five would force operators to invent profiles that must not exist, so the
+launcher accepts an explicitly validated subset.
+
+### Running a launch
+
+```bash
+python3 -m mcp_application_launcher \
+    --client claude_code \
+    --namespaces author,reviewer,merger \
+    --profile author=prgs-author \
+    --profile reviewer=prgs-reviewer \
+    --profile merger=prgs-merger
+```
+
+That mints one trusted `client_instance_id`, writes a per-launch `mcpServers`
+config, and executes:
+
+```text
+claude --mcp-config <per-launch.json> --strict-mcp-config
+```
+
+Add `--dry-run` to print the plan (namespaces, excluded namespaces, minted ID,
+argv, config path) as JSON and start nothing. Omit `--namespaces` to launch all
+five exactly as before. Arguments after the flags are forwarded to the client.
+
+Other supported clients register one entry in
+`mcp_application_launcher.CLIENT_LAUNCH_SPECS`; because the argv builder only
+ever receives a config this module wrote, every client necessarily goes through
+the same mint-once/propagate-to-all mechanism.
+
+### Why the config is per-launch and never `.mcp.json`
+
+Persisting a trusted ID into a shared, reused `.mcp.json` would give two
+concurrent sessions **the same** trusted identity — precisely the reuse case
+the duplicate gate must reject. The launcher therefore writes a fresh
+owner-readable-only (`0600`) config per launch. Do not commit one, and do not
+copy a minted `GITEA_MCP_CLIENT_INSTANCE` into any checked-in configuration.
+
+### Provenance sealing
+
+The `inst-…` format is public and reproducible, so format alone cannot
+establish trust: anyone able to set one environment variable could otherwise
+hand-write a valid-looking ID and be believed. Trust therefore requires **both**
+the format and `GITEA_MCP_INSTANCE_PROVENANCE=trusted_launcher`, which only the
+launcher writes. A well-formed but unsealed identity is classified
+`unsealed_launcher` and **fails closed** — it is still reported for diagnosis,
+but never authorizes trusted attribution. Manually asserted trust is not
+possible.
+
+### Migration and coordinated relaunch
+
+Existing configurations that predate this work set no instance key at all, so
+their workers register under `legacy-pid-*` (`legacy_incomplete`) and remain
+mutually indistinguishable when two launches share a profile.
+
+Migrating is a **coordinated relaunch**, not an in-place edit — a running
+worker cannot acquire an identity it was not started with:
+
+1. Stop every LLM client currently running Gitea MCP workers. A single
+   surviving legacy cohort keeps the fleet ambiguous.
+2. Relaunch each application through the command above.
+3. Repeat per application. Concurrent launches are expected and safe: each
+   receives its own trusted ID.
+
+### Post-launch verification
+
+* `gitea_get_runtime_context` → `provenance_assessment.attachment` should show
+  an `inst-…` `client_instance_id` with
+  `instance_id_provenance: trusted_launcher`, not `legacy-pid-*` /
+  `legacy_incomplete`.
+* Every namespace of the same launch must report that **same** ID; two
+  different launches must report different IDs.
+* `gitea_resolve_task_capability` should return
+  `exact_safe_next_action: "None; ready for operations."` with no
+  `blocker_kind`. A `runtime_reconnect_required` naming duplicate PIDs per
+  profile means at least one cohort is still on legacy identity, or a second
+  cohort is genuinely running.
+* `mcp_application_launcher.collect_instance_ids_from_mcp_servers(servers,
+  namespaces=[...])` returns `shared_single_trusted_id` for a built config, and
+  reports `missing_servers` when an expected namespace entry is absent.
+
+## How five workers join one instance
+
+1. The host starts one application instance (for example one Codex session).
+2. The **production application launcher**
+   (`mcp_application_launcher.build_application_mcp_servers` /
+   `gitea_config.multi_namespace_launcher_entries`) mints exactly one
+   `client_instance_id` via `mcp_fleet_snapshot.generate_client_instance_id`
+   and injects the same env into every namespace worker:
+
+   ```text
+   GITEA_MCP_CLIENT=<client_type>
+   GITEA_MCP_CLIENT_INSTANCE=<client_instance_id>
+   GITEA_MCP_INSTANCE_PROVENANCE=trusted_launcher
+   GITEA_MCP_CLIENT_SESSION=<session_id>          # optional but recommended
+   GITEA_MCP_FLEET_RUN_ID=<enrollment id>         # when on an approved canary
+   GITEA_CLIENT_MANAGED=1
+   GITEA_MCP_PROFILE=<role-profile>
+   ```
+
+3. The MCP client starts the five namespace processes (`gitea-author`,
+   `gitea-reviewer`, `gitea-merger`, `gitea-controller`, `gitea-reconciler`)
+   from that generated `mcpServers` map — each entry carries the **same**
+   instance ID.
+4. Each worker registers once into the worker registry with its own
+   `worker_identity`, `namespace`, `generation_id`, and PID, but the shared
+   `client_instance_id`. Workers never mint a trusted instance ID themselves.
+
+Only IDs matching the launcher format `inst-<client>-<timestamp>-<digest>`
+are trusted. Missing, `legacy-pid-*` / `pid-*` placeholders, and ordinary
+user-supplied strings fail closed as untrusted and cannot authorize
+multi-instance fleet mutation safety.
+
+Worker reconnect (same process, same registration) keeps the instance ID.
+Worker restart (new process) mints a new worker identity and generation but
+must still receive the same `GITEA_MCP_CLIENT_INSTANCE` from the parent
+application if it is the same launch. Full application restart mints a new
+`client_instance_id` (call the launcher without a prior ID).
+
+### Mutation gate (instance-aware)
+
+The capability / runtime diagnostic gate
+(`_check_mcp_runtimes_diagnostics`) is **instance-aware**:
+
+* Two legitimate application instances that share a profile (same
+  `GITEA_MCP_PROFILE`) are allowed when each has a distinct trusted
+  `client_instance_id`.
+* More than one live worker for the same
+  `(client_instance_id, profile/namespace)` fails closed as a duplicate
+  namespace worker.
+* Multiple processes sharing a profile **without** trusted instance
+  evidence still fail closed (indistinguishable from a duplicate).
+
+## Approved fleet manifest
+
+An operator (or controller enrollment step) obtains an approved manifest as a
+list of expected instances, for example:
+
+```json
+{
+  "instances": [
+    {
+      "client_type": "codex",
+      "client_instance_id": "inst-codex-20260730T120000Z-abc123def456",
+      "fleet_run_id": "canary-963-2026-07-30",
+      "namespaces": ["author", "reviewer", "merger", "controller", "reconciler"]
+    },
+    {
+      "client_type": "codex",
+      "client_instance_id": "inst-codex-20260730T120100Z-fed654cba321",
+      "fleet_run_id": "canary-963-2026-07-30",
+      "namespaces": ["author", "reviewer", "merger", "controller", "reconciler"]
+    }
+  ]
+}
+```
+
+Pass that list as `expected_manifest` to `gitea_snapshot_instance_fleet`.
+When a manifest is supplied:
+
+* instances on the manifest but not live → `missing_expected`;
+* live instances not on the manifest → `unmanifested`;
+* both are **active blockers** for fleet-gate safety.
+
+Without a manifest, the snapshot still enumerates the live fleet and classifies
+identity collisions; it does not invent enrollment policy.
+
+## Snapshot consistency
+
+Each snapshot includes:
+
+* `snapshot_at` — UTC timestamp;
+* `consistency_token` / `registry_revision` — content digest over worker
+  identity, instance id, status, heartbeat, fencing, and generation.
+
+Two successive snapshots can prove heartbeat continuity via
+`mcp_fleet_snapshot.compare_snapshot_heartbeats`.
+
+## Classification (active vs historical)
+
+**Active blockers** (make `live_fleet_safe=false`):
+
+* missing expected instance;
+* unmanifested extra instance;
+* duplicate namespace worker within one instance;
+* live `client_instance_id` collision;
+* reused worker / session / generation / process / PID / fencing identity;
+* orphaned or unowned workers;
+* unknown client;
+* foreign-repository workers;
+* old-revision workers;
+* stale workers still marked active;
+* legacy incomplete instance identity.
+
+**Historical dead rows** (`status` released/superseded) are reported as
+`historical_dead` findings with `active_blocker=false`. They never
+automatically make the live fleet unsafe.
+
+## Fail-closed behaviour and recovery
+
+| Condition | Mutation safety | Diagnostic reads | Recovery |
+| --- | --- | --- | --- |
+| Two instances, same `client_type`, distinct IDs | Safe (if otherwise healthy) | Available | None needed |
+| Live reuse of one `client_instance_id` | Unsafe | Available | Stop the colliding launch or re-issue a distinct ID |
+| Two workers, same namespace, one instance | Unsafe | Available | Stop the extra worker |
+| Legacy registration without trusted instance ID | Unsafe for fleet mutation | Available | Relaunch with launcher-issued `GITEA_MCP_CLIENT_INSTANCE` |
+| Historical dead row only | Does not block alone | Available | No action required for fleet safety |
+| Registry unavailable | Snapshot fails closed | N/A | Repair registry path / reconnect namespaces |
+
+Incomplete or untrusted instance identity **cannot** authorize unsafe
+mutation. Diagnostic reads remain available where `gitea.read` allows.
+
+## Permissions
+
+* **Allowed:** `controller`, `reconciler` with `gitea.read`.
+* **Denied:** author, reviewer, merger (even with `gitea.read` for other tools).
+* **No new mutation permissions** are granted to any role.
+
+## Related surfaces
+
+* `mcp_worker_identity` — registry, worker identity, heartbeats (#948, #975).
+* `mcp_fleet_snapshot` — pure snapshot + classification (#978).
+* `gitea_snapshot_instance_fleet` — sanctioned MCP tool (#978).
+* `gitea_get_runtime_context` — single-process view (not fleet-wide).
+
+## Non-goals
+
+* Starting the #963 canary.
+* Purging historical registry rows.
+* Preserving “exactly one process per profile” as the permanent model.
+* Using shell / process-table / SQLite inspection as production fleet evidence.

docs/mcp-tool-inventory.md

@@ -159,6 +159,7 @@ that gates each call, not which tools exist.
 - `gitea_sentry_reconcile_issue`
 - `gitea_sentry_watchdog`
 - `gitea_set_issue_labels`
+- `gitea_snapshot_instance_fleet`
 - `gitea_submit_pr_review`
 - `gitea_update_pr_branch_by_merge`
 - `gitea_validate_review_final_report`

gitea_config.py

@@ -1193,6 +1193,31 @@ RECOGNIZED_GITEA_ENV_KEYS = frozenset({
     "GITEA_IRRECOVERABLE_HMAC_SECRET",
     "GITEA_FORCE_MCP_RUNTIME_CHECK",
     "GITEA_FORCE_CLIENT_MANAGED",
+    # #975: the client-identity inputs the server actually consumes at startup
+    # (CLIENT_NAME_ENV / CLIENT_INSTANCE_ENV / CLIENT_SESSION_ENV in
+    # gitea_mcp_server). Production read them while this allowlist omitted them,
+    # so the peer-env scan classified them as unsupported overrides and the
+    # capability resolver refused every mutation fleet-wide. Named individually
+    # on purpose: no prefix is added, so an unrecognised GITEA_* override is
+    # still refused exactly as it was before.
+    "GITEA_MCP_CLIENT",
+    "GITEA_MCP_CLIENT_INSTANCE",
+    "GITEA_MCP_CLIENT_SESSION",
+    # #978: operator-approved fleet enrollment identity shared by one launch.
+    "GITEA_MCP_FLEET_RUN_ID",
+    "GITEA_MCP_PROCESS_IDENTITY",
+    # #978 B1/B2: launcher-sealed provenance + peer-visible worker identity so
+    # the instance-aware mutation gate can distinguish two legitimate
+    # application instances that share a profile from a true duplicate worker.
+    "GITEA_MCP_INSTANCE_PROVENANCE",
+    "GITEA_MCP_WORKER_IDENTITY",
+    "GITEA_MCP_GENERATION_ID",
+    # #975 review 652 B1: production also consumes HEARTBEAT_INTERVAL_ENV from
+    # mcp_worker_identity via gitea_mcp_server._start_worker_heartbeat. Omitting
+    # it reproduced the same unsupported-env → runtime_reconnect_required
+    # failure mode for the documented operator override. Named individually;
+    # no GITEA_* / GITEA_WORKER_* prefix is added.
+    "GITEA_WORKER_HEARTBEAT_INTERVAL_SECONDS",
 })
 
 RECOGNIZED_GITEA_ENV_PREFIXES = (
@@ -1220,24 +1245,74 @@ def get_unconsumed_gitea_env_overrides(env=None) -> dict[str, str]:
     return unconsumed
 
 
-def launcher_entry(profile_name, config_path=None):
+def launcher_entry(
+    profile_name,
+    config_path=None,
+    *,
+    client_type=None,
+    client_instance_id=None,
+    fleet_run_id=None,
+    session_id=None,
+    launch_nonce=None,
+):
     """Return a thin MCP launcher entry for *profile_name*.
 
-    Contains command/args and the GITEA_MCP_* / GITEA_CLIENT_MANAGED env vars — never a token
-    or password. Suitable for Claude / Gemini / Codex ``mcpServers`` blocks.
+    Contains command/args and the GITEA_MCP_* / GITEA_CLIENT_MANAGED env vars —
+    never a token or password. Suitable for Claude / Gemini / Codex
+    ``mcpServers`` blocks.
+
+    #978 B1: production launches mint (or reuse) a trusted
+    ``GITEA_MCP_CLIENT_INSTANCE`` so workers never fall back to the legacy
+    placeholder identity on the real serve path. Pass *client_instance_id* to
+    resume the same launch; omit it for a fresh launch (new ID).
     """
-    command, args = server_command()
-    return {
-        "gitea-tools": {
-            "command": command,
-            "args": args,
-            "env": {
-                "GITEA_MCP_CONFIG": config_path or DEFAULT_CONFIG_PATH,
-                "GITEA_MCP_PROFILE": profile_name,
-                "GITEA_CLIENT_MANAGED": "1",
-            },
-        }
-    }
+    import mcp_application_launcher as app_launcher
+
+    built = app_launcher.launcher_entry_for_profile(
+        profile_name,
+        client_type=client_type or "unknown",
+        config_path=config_path or DEFAULT_CONFIG_PATH,
+        client_instance_id=client_instance_id,
+        fleet_run_id=fleet_run_id,
+        session_id=session_id,
+        launch_nonce=launch_nonce,
+        server_key="gitea-tools",
+    )
+    # Public shape stays {server_key: {command, args, env}} for existing callers.
+    return {"gitea-tools": built["gitea-tools"]}
+
+
+def multi_namespace_launcher_entries(
+    profile_by_namespace,
+    *,
+    client_type,
+    namespaces=None,
+    config_path=None,
+    client_instance_id=None,
+    fleet_run_id=None,
+    session_id=None,
+    launch_nonce=None,
+):
+    """Build production mcpServers for the namespaces of one application launch.
+
+    One shared trusted ``client_instance_id`` is minted (or reused) and
+    propagated to every namespace worker env. *namespaces* selects a validated
+    project-scoped subset (#985); omitting it launches all five sanctioned
+    namespaces as before. See
+    :func:`mcp_application_launcher.build_application_mcp_servers`.
+    """
+    import mcp_application_launcher as app_launcher
+
+    return app_launcher.build_application_mcp_servers(
+        profile_by_namespace,
+        client_type=client_type,
+        namespaces=namespaces,
+        config_path=config_path or DEFAULT_CONFIG_PATH,
+        client_instance_id=client_instance_id,
+        fleet_run_id=fleet_run_id,
+        session_id=session_id,
+        launch_nonce=launch_nonce,
+    )
 
 
 

gitea_mcp_server.py

@@ -1052,9 +1052,16 @@ def _create_issue_bootstrap_assessment(
         git_state = issue_lock_worktree.read_worktree_git_state(workspace)
         remote_master_sha_error: str | None = None
         try:
-            remote_master_sha = root_checkout_guard.resolve_remote_master_sha(
+            # #983: consume the resolved-target state so an *underivable* base ref
+            # reaches the assessor as named missing evidence rather than a bare
+            # None, which the bootstrap would otherwise report only as "live
+            # master tip is unknown" with no cause.
+            _base_state = root_checkout_guard.resolve_remote_master_ref_state(
                 ctx["canonical_repo_root"]
             )
+            remote_master_sha = _base_state["sha"]
+            if not remote_master_sha and _base_state.get("reasons"):
+                remote_master_sha_error = "; ".join(_base_state["reasons"])
         except Exception as exc:
             remote_master_sha = None
             remote_master_sha_error = (
@@ -1082,9 +1089,14 @@ def _create_issue_bootstrap_assessment(
     # closed instead of proceeding without base-equivalence proof.
     remote_master_sha_error: str | None = None
     try:
-        remote_master_sha = root_checkout_guard.resolve_remote_master_sha(
+        # #983: same resolved-target consumption as the author bootstrap above —
+        # a target whose base ref cannot be derived must say why.
+        _base_state = root_checkout_guard.resolve_remote_master_ref_state(
             ctx["canonical_repo_root"]
         )
+        remote_master_sha = _base_state["sha"]
+        if not remote_master_sha and _base_state.get("reasons"):
+            remote_master_sha_error = "; ".join(_base_state["reasons"])
     except Exception as exc:
         remote_master_sha = None
         remote_master_sha_error = f"{type(exc).__name__}: {exc}".strip() or "resolver failed"
@@ -1303,7 +1315,12 @@ def _run_anti_stomp_preflight(
     workspace = ctx["workspace_path"]
     canonical_root = ctx["canonical_repo_root"]
     git_state = issue_lock_worktree.read_worktree_git_state(canonical_root)
-    remote_master_sha = root_checkout_guard.resolve_remote_master_sha(canonical_root)
+    # #983: derive the target base ref for this repository and carry the ref
+    # itself alongside the SHA, so the anti-stomp root-checkout check reports the
+    # ref it actually compared.
+    _root_base_state = root_checkout_guard.resolve_remote_master_ref_state(canonical_root)
+    remote_master_sha = _root_base_state["sha"]
+    remote_master_ref = _root_base_state.get("ref")
 
     # Repo/org facts (best-effort; explicit org/repo when provided).
     resolved_org = org
@@ -1432,6 +1449,7 @@ def _run_anti_stomp_preflight(
         root_head_sha=git_state.get("head_sha"),
         root_porcelain=git_state.get("porcelain_status") or "",
         remote_master_sha=remote_master_sha,
+        remote_master_ref=remote_master_ref,
         startup_head=startup_head,
         current_code_head=current_code_head,
         lease_required=lease_required,
@@ -1916,8 +1934,16 @@ def _verify_role_mutation_workspace(
             if runtime_reasons:
                 raise RuntimeError("; ".join(runtime_reasons))
     except Exception as exc:
-        if "stale-runtime:" in str(exc):
-            raise RuntimeError(str(exc))
+        # #975: ``_check_mcp_runtimes_diagnostics`` raises every reason it
+        # produces through this one RuntimeError, but only ``stale-runtime:``
+        # was re-raised here — an ``unsupported-env:`` reason was swallowed
+        # while still failing the capability resolver, so this preflight and
+        # the resolver disagreed about the identical diagnostic. Both
+        # authoritative prefixes now propagate the same way. This can only ever
+        # widen what is refused, never widen what is permitted.
+        message = str(exc)
+        if any(prefix in message for prefix in RUNTIME_DIAGNOSTIC_HARD_PREFIXES):
+            raise RuntimeError(message)
         pass
 
     role = _effective_workspace_role()
@@ -1979,7 +2005,11 @@ def _enforce_root_checkout_guard(worktree_path: str | None = None) -> None:
     canonical_root = ctx["canonical_repo_root"]
     workspace = ctx["workspace_path"]
     git_state = issue_lock_worktree.read_worktree_git_state(canonical_root)
-    remote_master_sha = root_checkout_guard.resolve_remote_master_sha(canonical_root)
+    # #983: consume the resolved target so the contamination message names the
+    # ref that was actually compared (refs/remotes/<remote>/<branch>) instead of
+    # a hardcoded 'prgs/master' the target repository may not have.
+    base_state = root_checkout_guard.resolve_remote_master_ref_state(canonical_root)
+    remote_master_sha = base_state["sha"]
     assessment = root_checkout_guard.assess_root_checkout_guard(
         workspace_path=workspace,
         canonical_repo_root=canonical_root,
@@ -1989,6 +2019,7 @@ def _enforce_root_checkout_guard(worktree_path: str | None = None) -> None:
         remote_master_sha=remote_master_sha,
         resolved_role=_preflight_resolved_role,
         actual_role=_actual_profile_role(),
+        remote_master_ref=base_state.get("ref"),
     )
     if assessment["block"]:
         raise RuntimeError(root_checkout_guard.format_root_checkout_guard_error(assessment))
@@ -15714,6 +15745,10 @@ _WORKER_REGISTRY = None
 _WORKER_IDENTITY: str | None = None
 _WORKER_GENERATION: str | None = None
 _WORKER_REGISTRATION_ATTEMPTED = False
+#: #975: the one heartbeat supervisor for this process's registration. One
+#: process registers exactly one worker identity, so there is exactly one
+#: supervisor and it is never replaced.
+_WORKER_HEARTBEAT_SUPERVISOR = None
 
 #: Env a client launcher may set to name itself and its session. Absent values
 #: are reported as unknown; they are never guessed at, because guessing is what
@@ -15721,6 +15756,10 @@ _WORKER_REGISTRATION_ATTEMPTED = False
 CLIENT_NAME_ENV = "GITEA_MCP_CLIENT"
 CLIENT_INSTANCE_ENV = "GITEA_MCP_CLIENT_INSTANCE"
 CLIENT_SESSION_ENV = "GITEA_MCP_CLIENT_SESSION"
+FLEET_RUN_ENV = "GITEA_MCP_FLEET_RUN_ID"
+INSTANCE_PROVENANCE_ENV = "GITEA_MCP_INSTANCE_PROVENANCE"
+WORKER_IDENTITY_ENV = "GITEA_MCP_WORKER_IDENTITY"
+GENERATION_ID_ENV = "GITEA_MCP_GENERATION_ID"
 
 
 def _worker_registry():
@@ -15744,13 +15783,57 @@ def _worker_registry():
 
 
 def _client_identity_hints() -> dict:
-    """What the launcher told us about itself. Unset fields stay unset."""
+    """What the launcher told us about itself (#948 / #975 / #978).
+
+    ``client_instance_id`` is established by the trusted application launcher
+    once per launch and shared by all five namespace workers. Workers inherit
+    that value from the production launcher env and never mint a trusted ID
+    themselves. When the launcher key is absent or untrusted we still register
+    under an explicit *legacy* placeholder so diagnostic reads work, but the
+    registration is marked incomplete and cannot authorize multi-instance fleet
+    mutation safety.
+    """
+    import mcp_application_launcher as app_launcher
+    import mcp_fleet_snapshot
+
+    inherited = app_launcher.inherit_or_refuse_client_instance(os.environ)
+    client_name = (os.environ.get(CLIENT_NAME_ENV) or "").strip() or None
+    instance = {
+        "client_instance_id": inherited.get("client_instance_id"),
+        "provenance": inherited.get("provenance"),
+        "trusted": bool(inherited.get("trusted")),
+        "complete": bool(inherited.get("complete")),
+        "launcher_sealed": bool(inherited.get("launcher_sealed")),
+    }
+    # Legacy placeholder only when the launcher omitted a usable key — never
+    # invent a trusted ID from PID proximity or ordinary user env.
+    if not instance["client_instance_id"] or not instance["trusted"]:
+        if not instance["client_instance_id"]:
+            placeholder = f"legacy-pid-{os.getpid()}"
+            fallback = mcp_fleet_snapshot.assess_instance_identity(
+                placeholder, source="pid_fallback"
+            )
+            instance = {
+                "client_instance_id": fallback["client_instance_id"],
+                "provenance": fallback["provenance"],
+                "trusted": False,
+                "complete": False,
+                "launcher_sealed": False,
+            }
+        else:
+            # Malformed / untrusted user-supplied value: keep it for diagnosis
+            # but never mark trusted.
+            instance["trusted"] = False
+            instance["launcher_sealed"] = False
     return {
-        "client_name": (os.environ.get(CLIENT_NAME_ENV) or "").strip() or None,
-        "client_instance_id": (os.environ.get(CLIENT_INSTANCE_ENV) or "").strip()
-        or f"pid-{os.getpid()}",
+        "client_name": client_name,
+        "client_instance_id": instance["client_instance_id"],
+        "instance_id_provenance": instance["provenance"],
+        "instance_identity_trusted": instance["trusted"],
+        "instance_launcher_sealed": instance.get("launcher_sealed", False),
         "session_id": (os.environ.get(CLIENT_SESSION_ENV) or "").strip()
         or f"proc-{os.getpid()}-{_process_boot_head_sha or 'nohead'}",
+        "fleet_run_id": (os.environ.get(FLEET_RUN_ENV) or "").strip() or None,
     }
 
 
@@ -15782,24 +15865,71 @@ def _active_worker_identity() -> str | None:
             identity = mcp_worker_identity.generate_worker_identity(
                 hints["client_name"], hints["session_id"]
             )
+            role = _active_role_kind_safe()
+            profile_name = (
+                (os.environ.get(gitea_config.ENV_PROFILE) or "").strip() or None
+            )
+            # Namespace is the role namespace (author/reviewer/…) when known.
+            namespace = role if role in {
+                "author", "reviewer", "merger", "controller", "reconciler"
+            } else None
+            parity = None
+            try:
+                parity = _current_master_parity()
+            except Exception:
+                parity = None
             outcome = registry.register(
                 worker_identity=identity,
                 client_name=hints["client_name"],
                 client_instance_id=hints["client_instance_id"],
                 session_id=hints["session_id"],
                 generation_id=generation,
-                role=_active_role_kind_safe(),
-                profile=(os.environ.get(gitea_config.ENV_PROFILE) or "").strip() or None,
+                role=role,
+                profile=profile_name,
+                namespace=namespace,
                 remote=(os.environ.get("GITEA_MCP_REMOTE") or "").strip() or None,
                 repository_binding=PROJECT_ROOT,
                 pid=os.getpid(),
                 transport=native.get("bound_transport"),
                 token_fingerprint=native.get("token_fingerprint"),
                 pid_alive_probe=issue_lock_store.is_process_alive,
+                fleet_run_id=hints.get("fleet_run_id"),
+                process_identity=f"pid-{os.getpid()}",
+                startup_revision=(parity or {}).get("startup_head")
+                or _process_boot_head_sha,
+                loaded_revision=(parity or {}).get("local_head")
+                or (parity or {}).get("current_head"),
+                parity_revision=(parity or {}).get("current_head"),
+                live_revision=(parity or {}).get("live_remote_head"),
+                instance_id_provenance=hints.get("instance_id_provenance"),
             )
             if outcome.get("registered"):
                 _WORKER_IDENTITY = identity
                 _WORKER_GENERATION = generation
+                # #978 B2: export worker/generation/instance into this process
+                # env so peer process scans (and the instance-aware mutation
+                # gate) can attribute live workers without treating shared
+                # profiles as duplicates. Never overwrite a launcher-sealed
+                # client_instance_id with a different value.
+                os.environ[WORKER_IDENTITY_ENV] = identity
+                os.environ[GENERATION_ID_ENV] = generation
+                if hints.get("client_instance_id") and not (
+                    os.environ.get(CLIENT_INSTANCE_ENV) or ""
+                ).strip():
+                    os.environ[CLIENT_INSTANCE_ENV] = str(
+                        hints["client_instance_id"]
+                    )
+                # #975: registration is the only moment identity and fencing
+                # epoch are both known, so the heartbeat supervisor is started
+                # here. Without it ``last_heartbeat_at`` never left
+                # ``started_at`` and every healthy client lost ownership at the
+                # TTL.
+                _start_worker_heartbeat(
+                    identity=identity,
+                    fencing_epoch=outcome.get("fencing_epoch"),
+                    hints=hints,
+                    generation=generation,
+                )
                 return identity
             if not outcome.get("collision"):
                 return None
@@ -15810,6 +15940,78 @@ def _active_worker_identity() -> str | None:
     return None
 
 
+def _start_worker_heartbeat(
+    *,
+    identity: str,
+    fencing_epoch,
+    hints: dict,
+    generation: str | None,
+) -> None:
+    """Attach a heartbeat supervisor to the registration just created (#975).
+
+    Never raises: a supervisor that cannot start leaves the row exactly as
+    ``register()`` wrote it, which is the pre-#975 behaviour, rather than
+    failing the tool call that happened to trigger lazy registration.
+
+    Under pytest the thread is deliberately not started. Tests drive
+    ``WorkerHeartbeatSupervisor`` directly with an injected clock, so the suite
+    proves the lifecycle without leaving background sqlite writers behind.
+    """
+    global _WORKER_HEARTBEAT_SUPERVISOR
+    if _WORKER_HEARTBEAT_SUPERVISOR is not None:
+        return
+    registry = _worker_registry()
+    if registry is None or fencing_epoch is None:
+        return
+    try:
+        supervisor = mcp_worker_identity.WorkerHeartbeatSupervisor(
+            registry,
+            worker_identity=identity,
+            fencing_epoch=int(fencing_epoch),
+            session_id=hints.get("session_id"),
+            generation_id=generation,
+            client_name=hints.get("client_name"),
+            pid=os.getpid(),
+            ttl_seconds=mcp_worker_identity.DEFAULT_HEARTBEAT_TTL_SECONDS,
+            interval_seconds=os.environ.get(
+                mcp_worker_identity.HEARTBEAT_INTERVAL_ENV
+            ),
+        )
+        _WORKER_HEARTBEAT_SUPERVISOR = supervisor
+        if not mcp_daemon_guard.is_pytest_runtime():
+            supervisor.start()
+    except Exception:
+        return
+
+
+def _worker_heartbeat_status() -> dict:
+    """Read-only heartbeat observability for ``gitea_get_runtime_context`` (#975).
+
+    Reports the unsupervised case explicitly rather than omitting the key, so an
+    operator can tell "no supervisor" apart from "supervisor with no beats yet".
+    """
+    supervisor = _WORKER_HEARTBEAT_SUPERVISOR
+    if supervisor is None:
+        return {
+            "supervised": False,
+            "running": False,
+            "reasons": [
+                "no worker heartbeat supervisor is attached to this process; the "
+                "registration is not being renewed and will go stale at its TTL"
+            ],
+        }
+    try:
+        status = dict(supervisor.status())
+    except Exception as exc:
+        return {
+            "supervised": True,
+            "running": False,
+            "reasons": [f"heartbeat status unavailable: {type(exc).__name__}: {exc}"],
+        }
+    status.setdefault("reasons", [])
+    return status
+
+
 def _active_role_kind_safe() -> str | None:
     """Best-effort role for the registry record; never raises into a tool call.
 
@@ -19278,6 +19480,172 @@ def gitea_validate_review_final_report(
     )
 
 
+@mcp.tool()
+def gitea_snapshot_instance_fleet(
+    remote: str = "dadeschools",
+    host: str | None = None,
+    org: str | None = None,
+    repo: str | None = None,
+    expected_manifest: list | dict | str | None = None,
+    include_historical: bool = True,
+    canonical_repository: str | None = None,
+    expected_live_revision: str | None = None,
+) -> dict:
+    """Read-only: authoritative instance-level fleet identity and health snapshot (#978).
+
+    Controller and reconciler only. Returns a point-in-time snapshot of every
+    registered namespace worker with instance attribution, heartbeat freshness,
+    and structured classification (expected/missing/unmanifested, duplicate
+    namespace workers, identity collisions, foreign repository, old revision,
+    historical dead rows). Sharing only a client type is never a duplicate.
+
+    Does not grant any mutation capability. Does not scan process tables or
+    open foreign databases for production evidence — the worker registry is
+    the sole authority.
+
+    Args:
+        remote: Known instance — 'dadeschools' or 'prgs'.
+        host: Optional host override.
+        org: Optional org override (audit context only).
+        repo: Optional repo override (audit context only).
+        expected_manifest: Optional approved fleet enrollment list
+            (list of dicts with client_instance_id / client_type / fleet_run_id).
+            When supplied, missing and unmanifested instances are classified.
+        include_historical: Include released/superseded rows (default True);
+            historical rows never make the live fleet unsafe by themselves.
+        canonical_repository: Expected repository binding path/slug for
+            foreign-repository classification.
+        expected_live_revision: When set, workers whose recorded revisions
+            differ are classified as old-revision.
+    """
+    import json as _json
+
+    import mcp_fleet_snapshot
+
+    read_block = _profile_operation_gate("gitea.read")
+    if read_block:
+        return {
+            "success": False,
+            "read_only": True,
+            "reasons": read_block,
+            "permission_report": _permission_block_report("gitea.read"),
+        }
+
+    profile = get_profile()
+    role = _profile_role_kind(profile)
+    if role not in {"controller", "reconciler"}:
+        return {
+            "success": False,
+            "read_only": True,
+            "allowed": False,
+            "denied_role": role,
+            "required_roles": ["controller", "reconciler"],
+            "mutation_performed": False,
+            "reasons": [
+                f"gitea_snapshot_instance_fleet is restricted to controller and "
+                f"reconciler roles; active role_kind is {role!r}. Author, "
+                "reviewer, and merger profiles keep gitea.read for diagnosis "
+                "elsewhere but do not receive this fleet surface, and no "
+                "unrelated mutation permission is granted."
+            ],
+            "exact_next_action": (
+                "Re-run from a prgs-controller or prgs-reconciler namespace."
+            ),
+        }
+
+    manifest: list | None = None
+    if expected_manifest is not None:
+        raw = expected_manifest
+        if isinstance(raw, str):
+            try:
+                raw = _json.loads(raw)
+            except Exception as exc:
+                return {
+                    "success": False,
+                    "read_only": True,
+                    "reasons": [
+                        f"expected_manifest is not valid JSON: {type(exc).__name__}"
+                    ],
+                }
+        if isinstance(raw, dict):
+            # Accept {"instances": [...]} or a single instance object.
+            if "instances" in raw and isinstance(raw["instances"], list):
+                raw = raw["instances"]
+            else:
+                raw = [raw]
+        if not isinstance(raw, list):
+            return {
+                "success": False,
+                "read_only": True,
+                "reasons": ["expected_manifest must be a list of instance objects"],
+            }
+        manifest = raw
+
+    registry = _worker_registry()
+    if registry is None:
+        return {
+            "success": False,
+            "read_only": True,
+            "reasons": [
+                "worker registry is unavailable; cannot produce an authoritative "
+                "fleet snapshot (fail closed)"
+            ],
+            "exact_next_action": (
+                "Ensure GITEA_WORKER_REGISTRY_DB is writable and re-run after "
+                "workers have registered."
+            ),
+        }
+
+    try:
+        if include_historical:
+            records = registry.list_workers(status=None)
+        else:
+            records = registry.list_workers(status=mcp_worker_identity.STATUS_ACTIVE)
+    except Exception as exc:
+        return {
+            "success": False,
+            "read_only": True,
+            "reasons": [
+                f"failed to read worker registry: {type(exc).__name__}: {_redact(str(exc))}"
+            ],
+        }
+
+    parity = None
+    try:
+        parity = _current_master_parity()
+    except Exception:
+        parity = None
+    live_rev = expected_live_revision or (parity or {}).get("live_remote_head")
+    canon = canonical_repository or PROJECT_ROOT
+
+    snapshot = mcp_fleet_snapshot.snapshot_instance_fleet(
+        records,
+        expected_manifest=manifest,
+        pid_alive_probe=issue_lock_store.is_process_alive,
+        canonical_repository=canon,
+        expected_live_revision=live_rev,
+    )
+    snapshot["role_kind"] = role
+    snapshot["profile"] = profile.get("profile_name")
+    snapshot["remote"] = _effective_remote(remote)
+    snapshot["repository"] = {
+        "org": org,
+        "repo": repo,
+        "canonical_repository": canon,
+    }
+    snapshot["mutation_performed"] = False
+    snapshot["permission_scope"] = {
+        "read_only": True,
+        "granted_operations": ["gitea.read"],
+        "denied_unrelated_mutations": True,
+        "note": (
+            "This capability is strictly observational. It does not authorize "
+            "branch, issue, PR, review, merge, or restart mutations."
+        ),
+    }
+    return snapshot
+
+
 @mcp.tool()
 def gitea_get_runtime_context(
     remote: str = "dadeschools",
@@ -19451,6 +19819,11 @@ def gitea_get_runtime_context(
         "fencing_epoch": provenance_assessment["fencing_epoch"],
         "conflicting_live_sessions": provenance_assessment["conflicting_live_sessions"],
         "provenance_assessment": provenance_assessment,
+        # #975: whether this registration is actually being renewed. Read-only,
+        # and it grants nothing — ownership still comes from the attachment
+        # record above. It exists so "my heartbeat stopped" is diagnosable
+        # before the TTL turns it into session_attachment_missing.
+        "worker_heartbeat": _worker_heartbeat_status(),
         "unconsumed_gitea_env": unconsumed_env,
         "preflight_ready": preflight["preflight_ready"],
         "preflight_block_reasons": preflight["preflight_block_reasons"],
@@ -21959,6 +22332,17 @@ def gitea_route_task_session(
 # self-recovery was removed from the read-only path (was _trigger_mcp_auto_restart).
 # Recovery is owned exclusively by the IDE/client reconnect path.
 
+#: Every reason prefix ``_check_mcp_runtimes_diagnostics`` can emit. Callers
+#: raise its reasons as one RuntimeError, and #975 found that the preflight
+#: re-raise recognised only ``stale-runtime:``, silently dropping
+#: ``unsupported-env:`` while the capability resolver still failed on it. This
+#: lives beside the producer so a newly added reason family cannot be forgotten
+#: by a distant re-raise predicate again.
+RUNTIME_DIAGNOSTIC_HARD_PREFIXES: tuple[str, ...] = (
+    "stale-runtime:",
+    "unsupported-env:",
+)
+
 
 def _check_mcp_runtimes_diagnostics(task: str, matching_profiles: list[str]) -> list[str]:
     """Read-only: report missing or stale MCP runtimes (no config or process mutation).
@@ -22053,6 +22437,20 @@ def _check_mcp_runtimes_diagnostics(task: str, matching_profiles: list[str]) ->
         )
         peer_worker_identity = peer_env.get("GITEA_MCP_WORKER_IDENTITY")
         peer_generation = peer_env.get("GITEA_MCP_GENERATION_ID")
+        # #978 B2: instance attribution for the live fleet gate. Two legitimate
+        # application instances may share a profile when each has a distinct
+        # trusted client_instance_id; only same-(instance, profile/namespace)
+        # duplicates remain blocked.
+        import mcp_fleet_snapshot as _fleet
+
+        peer_instance_raw = peer_env.get("GITEA_MCP_CLIENT_INSTANCE")
+        peer_instance_assess = _fleet.assess_instance_identity(peer_instance_raw)
+        peer_client_instance = (
+            peer_instance_assess["client_instance_id"]
+            if peer_instance_assess.get("trusted")
+            else None
+        )
+        peer_instance_trusted = bool(peer_instance_assess.get("trusted"))
 
         for env_match in re.finditer(r'\b(GITEA_[A-Z0-9_]+)=([^\s]+)', env_out):
             k, v = env_match.group(1), env_match.group(2)
@@ -22070,6 +22468,9 @@ def _check_mcp_runtimes_diagnostics(task: str, matching_profiles: list[str]) ->
             "is_client_managed": is_client_managed,
             "worker_identity": peer_worker_identity,
             "generation_id": peer_generation,
+            "client_instance_id": peer_client_instance,
+            "instance_identity_trusted": peer_instance_trusted,
+            "raw_client_instance_id": peer_instance_raw,
         }
         if profile not in all_profile_procs:
             all_profile_procs[profile] = []
@@ -22077,44 +22478,85 @@ def _check_mcp_runtimes_diagnostics(task: str, matching_profiles: list[str]) ->
 
     running_profiles = {}
     for profile, procs in all_profile_procs.items():
-        # #948 AC40/AC43: sharing a profile is legitimate — profile is a
-        # reusable capability definition, not a worker identity. What is *not*
-        # legitimate is reusing one worker identity, or two live sessions
-        # claiming one generation. Distinctness has to be proven, though:
-        # processes carrying no identity evidence are indistinguishable, so
-        # they stay classified as duplicates and keep the #686 wall intact.
-        identified = [p for p in procs if p.get("worker_identity")]
-        distinct_identities = {p["worker_identity"] for p in identified}
-        all_identified = len(identified) == len(procs)
-        duplicate_identity = len(identified) != len(distinct_identities)
-        contested_generation = any(
-            len({p["worker_identity"] for p in identified if p.get("generation_id") == gen}) > 1
-            for gen in {p.get("generation_id") for p in identified if p.get("generation_id")}
-        )
-
-        if len(procs) > 1 and (
-            not all_identified or duplicate_identity or contested_generation
-        ):
-            pids_str = ", ".join(str(p["pid"]) for p in procs)
-            if duplicate_identity or contested_generation:
-                detail = (
-                    "The same worker identity or generation is claimed more than once, "
-                    "so these are genuine duplicates rather than independent workers."
-                )
+        # #978 B2 / #948 AC40: sharing a profile is legitimate when each live
+        # process belongs to a distinct trusted application instance. The
+        # permanent model is instance-aware — not exactly_one_per_profile.
+        # Fail closed only when:
+        #   * more than one live worker claims the same (client_instance_id,
+        #     profile/namespace), or
+        #   * worker identity / generation is reused, or
+        #   * multiple processes share a profile without trusted instance
+        #     evidence (indistinguishable → treat as duplicate).
+        # Group by trusted client_instance_id (untrusted/missing → one bucket).
+        by_instance: dict[str, list[dict]] = {}
+        for p in procs:
+            if p.get("instance_identity_trusted") and p.get("client_instance_id"):
+                key = str(p["client_instance_id"])
             else:
-                detail = (
-                    "Manual or duplicate launches defeat staleness detection and cannot "
-                    "receive client stdio."
+                key = "__untrusted_or_missing__"
+            by_instance.setdefault(key, []).append(p)
+
+        for instance_key, group in by_instance.items():
+            if len(group) <= 1:
+                continue
+            identified = [p for p in group if p.get("worker_identity")]
+            distinct_identities = {p["worker_identity"] for p in identified}
+            all_identified = len(identified) == len(group)
+            duplicate_identity = len(identified) != len(distinct_identities)
+            contested_generation = any(
+                len(
+                    {
+                        p["worker_identity"]
+                        for p in identified
+                        if p.get("generation_id") == gen
+                    }
                 )
-            reasons.append(
-                f"stale-runtime: Duplicate MCP server process(es) detected for profile '{profile}' (PIDs: {pids_str}). "
-                + detail
+                > 1
+                for gen in {
+                    p.get("generation_id")
+                    for p in identified
+                    if p.get("generation_id")
+                }
             )
-        # Otherwise: several independently identified workers share one profile.
-        # No reason is appended, deliberately. Every reason this function
-        # returns is raised as a hard RuntimeError by its callers, so recording
-        # legitimate concurrency here as "informational" would block exactly the
-        # case #948 exists to permit.
+            # Same trusted (instance, profile) with >1 live process is always a
+            # duplicate-namespace-worker, even if worker identities differ —
+            # one application launch gets exactly one worker per namespace.
+            same_trusted_instance = instance_key != "__untrusted_or_missing__"
+            if same_trusted_instance or (
+                not all_identified or duplicate_identity or contested_generation
+            ):
+                pids_str = ", ".join(str(p["pid"]) for p in group)
+                if same_trusted_instance:
+                    detail = (
+                        f"More than one live worker for profile '{profile}' under "
+                        f"client_instance_id {instance_key!r} (duplicate namespace "
+                        "worker). Two legitimate application instances with "
+                        "distinct trusted client_instance_id values may share a "
+                        "profile; this collision does not."
+                    )
+                elif duplicate_identity or contested_generation:
+                    detail = (
+                        "The same worker identity or generation is claimed more "
+                        "than once, so these are genuine duplicates rather than "
+                        "independent workers."
+                    )
+                else:
+                    detail = (
+                        "Multiple processes share this profile without distinct "
+                        "trusted client_instance_id evidence, so they cannot be "
+                        "told apart from a duplicate MCP process. Relaunch via "
+                        "the production application launcher so each instance "
+                        "receives a trusted GITEA_MCP_CLIENT_INSTANCE."
+                    )
+                reasons.append(
+                    f"stale-runtime: Duplicate MCP server process(es) detected "
+                    f"for profile '{profile}' (PIDs: {pids_str}). "
+                    + detail
+                )
+        # Multiple trusted instances sharing one profile intentionally produce
+        # no reason here. Callers raise every reason as a hard RuntimeError, so
+        # treating legitimate multi-instance concurrency as "informational"
+        # would re-introduce the profile-only wall #978 removes.
         client_procs = [p for p in procs if p["is_client_managed"]]
         if client_procs:
             client_procs.sort(key=lambda p: p["start_time"], reverse=True)

master_parity_gate.py

@@ -378,6 +378,11 @@ def format_parity(assessment: dict) -> str:
 # feeds the mutation gate and never changes startup_head/current_head.
 # ---------------------------------------------------------------------------
 
+# Legacy hardcoded target tracking ref. Retained for callers that still pass an
+# explicit ref, but no longer the default: assuming a remote named ``origin``
+# read an unrelated (often orphaned) remote-tracking ref in any checkout whose
+# remote is named something else, and reported the target stale against a commit
+# from a remote that may no longer even be configured (#983).
 DEFAULT_TARGET_TRACKING_REF = "refs/remotes/origin/master"
 
 
@@ -408,7 +413,7 @@ def assess_target_repository_parity(
     *,
     canonical_root: str | None,
     source: str | None,
-    tracking_ref: str = DEFAULT_TARGET_TRACKING_REF,
+    tracking_ref: str | None = None,
 ) -> dict:
     """Assess the configured cross-repository target checkout.
 
@@ -421,6 +426,11 @@ def assess_target_repository_parity(
     An unconfigured namespace is ``configured=False`` and never ``stale`` — the
     single-repository default has no second dimension to be stale about. A
     configured root that cannot be read is ``determinable=False`` with reasons.
+
+    *tracking_ref* defaults to None, meaning **derive the target from the
+    repository itself** through the same resolver the mutation guard uses, so
+    gating and reporting can never disagree about which ref is authoritative
+    (#983). Passing an explicit ref preserves the previous behaviour verbatim.
     """
     result = {
         "configured": bool(canonical_root),
@@ -429,9 +439,13 @@ def assess_target_repository_parity(
         "repository_slug": None,
         "checkout_head": None,
         "tracking_ref": tracking_ref,
+        "base_remote": None,
+        "base_branch": None,
+        "tracking_ref_source": "explicit_tracking_ref" if tracking_ref else None,
         "remote_tracking_head": None,
         "determinable": False,
         "stale": False,
+        "reason_code": None,
         "reasons": [],
     }
     if not canonical_root:
@@ -463,18 +477,53 @@ def assess_target_repository_parity(
     result["checkout_head"] = head
     result["determinable"] = True
 
-    remote_url = _git_capture(canonical_root, "remote", "get-url", "origin")
-    if remote_url:
-        # Local import keeps this module dependency-light for its startup role.
-        import remote_repo_guard
+    # Local import keeps this module dependency-light for its startup role.
+    import canonical_repository_root as _crr
 
-        parsed = remote_repo_guard.parse_org_repo_from_remote_url(remote_url)
-        if parsed:
-            result["repository_slug"] = f"{parsed[0]}/{parsed[1]}"
-    if not result["repository_slug"]:
-        result["reasons"].append(
-            "target repository identity could not be derived from its git remote"
+    # #983: identity comes from whichever remote actually proves it, not from a
+    # remote assumed to be named 'origin'. In a checkout whose only remote is
+    # 'prgs', the old lookup failed outright and reported the identity as
+    # underivable while a leftover refs/remotes/origin/master still resolved.
+    #
+    # #983 B2: this must be the *ambiguity-aware* resolver. The first-wins
+    # `resolve_identity_remote` picks whichever remote probes first, so on a
+    # target where distinct remotes claim different repositories the report
+    # confidently named one of them while the mutation gate refused the same
+    # target — gating and reporting evaluating different repositories, which is
+    # precisely the divergence this issue exists to end.
+    identity = _crr.assess_identity_remote(canonical_root)
+    result["repository_slug"] = identity["slug"]
+    if not identity["slug"]:
+        result["reason_code"] = identity["reason_code"]
+        result["reasons"].extend(
+            identity["reasons"]
+            or ["target repository identity could not be derived from its git remote"]
         )
+    if identity["ambiguous"]:
+        # An ambiguous target has no single authoritative base, so reporting one
+        # would be a guess. Fail closed here exactly as the gate does.
+        return result
+
+    # Identity is otherwise resolved independently of the base ref: a target that
+    # has never been fetched still has a provable repository identity, and
+    # reporting it as unidentifiable would lose real information over an
+    # unrelated missing ref.
+    if not tracking_ref:
+        # No remote argument. The identity remote resolved just above was
+        # *inferred here*, and feeding it back in would tell the resolver a
+        # caller had explicitly disambiguated the repository, suppressing its
+        # ambiguity gate (#983 B2). Only an operator-supplied remote may do that,
+        # and this call site has none.
+        base = _crr.resolve_target_base_ref(canonical_root)
+        if not base.get("proven"):
+            result["reason_code"] = base.get("reason_code")
+            result["reasons"].extend(base.get("reasons") or [])
+            return result
+        tracking_ref = base["tracking_ref"]
+        result["tracking_ref"] = tracking_ref
+        result["tracking_ref_source"] = base.get("source")
+        result["base_remote"] = base.get("remote")
+        result["base_branch"] = base.get("branch")
 
     tracking_head = _git_capture(canonical_root, "rev-parse", tracking_ref)
     if not tracking_head:

mcp_application_launcher.py

@@ -0,0 +1,753 @@
+"""Production application launcher for multi-namespace MCP fleets (#978 B1).
+
+One real LLM application launch mints exactly one trusted
+``client_instance_id`` and propagates it to every Gitea MCP namespace worker
+started for that launch. Workers never invent a trusted instance identity from
+PID proximity, timestamps, or ordinary untrusted environment values.
+
+This module is the production serve-path authority for instance identity.
+Tests and fixtures may call the same functions, but production registration
+receives the identity from the env this launcher builds — not from a hand-set
+test-only helper that bypasses it.
+"""
+
+from __future__ import annotations
+
+import os
+import secrets
+from typing import Any, Mapping, Sequence
+
+import mcp_fleet_snapshot as fleet
+import mcp_worker_identity as mwi
+
+#: Canonical MCP server names for the five role namespaces.
+NAMESPACE_SERVER_NAMES: tuple[str, ...] = (
+    "gitea-author",
+    "gitea-reviewer",
+    "gitea-merger",
+    "gitea-controller",
+    "gitea-reconciler",
+)
+
+#: Map MCP server name → role/namespace kind.
+SERVER_TO_NAMESPACE: dict[str, str] = {
+    "gitea-author": "author",
+    "gitea-reviewer": "reviewer",
+    "gitea-merger": "merger",
+    "gitea-controller": "controller",
+    "gitea-reconciler": "reconciler",
+}
+
+SANCTIONED_NAMESPACES: tuple[str, ...] = (
+    "author",
+    "reviewer",
+    "merger",
+    "controller",
+    "reconciler",
+)
+
+# Env the launcher may set on every worker of one application launch.
+CLIENT_NAME_ENV = "GITEA_MCP_CLIENT"
+CLIENT_INSTANCE_ENV = fleet.CLIENT_INSTANCE_ENV
+FLEET_RUN_ENV = fleet.FLEET_RUN_ENV
+CLIENT_SESSION_ENV = "GITEA_MCP_CLIENT_SESSION"
+CLIENT_MANAGED_ENV = "GITEA_CLIENT_MANAGED"
+PROFILE_ENV = "GITEA_MCP_PROFILE"
+CONFIG_ENV = "GITEA_MCP_CONFIG"
+INSTANCE_PROVENANCE_ENV = "GITEA_MCP_INSTANCE_PROVENANCE"
+WORKER_IDENTITY_ENV = "GITEA_MCP_WORKER_IDENTITY"
+GENERATION_ID_ENV = "GITEA_MCP_GENERATION_ID"
+
+# Marker the trusted launcher alone writes; ordinary user env without this
+# marker is never classified as launcher-trusted provenance.
+LAUNCHER_PROVENANCE_VALUE = fleet.INSTANCE_ID_PROVENANCE_TRUSTED
+
+
+def mint_application_launch(
+    client_type: str | None,
+    *,
+    launch_nonce: str | None = None,
+    fleet_run_id: str | None = None,
+    session_id: str | None = None,
+    now=None,
+) -> dict[str, Any]:
+    """Mint one trusted application-instance identity for a production launch.
+
+    Called exactly once per real application launch. The returned
+    ``client_instance_id`` is injected into every namespace worker environment
+    for that launch. A second call (separate launch) yields a different ID.
+    """
+    client = mwi.normalize_client_name(client_type)
+    instance_id = fleet.generate_client_instance_id(
+        client, launch_nonce=launch_nonce, now=now
+    )
+    assessment = fleet.assess_instance_identity(instance_id)
+    if not assessment["trusted"]:
+        # generate_client_instance_id always produces a trusted format; fail
+        # closed if that invariant ever breaks rather than shipping untrusted.
+        raise RuntimeError(
+            f"launcher produced untrusted client_instance_id {instance_id!r}: "
+            f"{assessment.get('reasons')}"
+        )
+    session = (session_id or "").strip() or f"launch-{secrets.token_hex(12)}"
+    return {
+        "client_type": client,
+        "client_instance_id": instance_id,
+        "instance_id_provenance": LAUNCHER_PROVENANCE_VALUE,
+        "instance_identity_trusted": True,
+        "fleet_run_id": (fleet_run_id or "").strip() or None,
+        "session_id": session,
+        "namespaces": list(SANCTIONED_NAMESPACES),
+        "namespace_server_names": list(NAMESPACE_SERVER_NAMES),
+    }
+
+
+def namespace_worker_env(
+    *,
+    profile_name: str,
+    client_type: str | None,
+    client_instance_id: str,
+    config_path: str | None = None,
+    fleet_run_id: str | None = None,
+    session_id: str | None = None,
+    extra_env: Mapping[str, str] | None = None,
+) -> dict[str, str]:
+    """Build the environment for one namespace worker of a trusted launch.
+
+    Never invents a client_instance_id. The caller must supply the launch-minted
+    identity so all five workers receive the same value.
+    """
+    assessment = fleet.assess_instance_identity(client_instance_id)
+    if not assessment["trusted"]:
+        raise ValueError(
+            "namespace_worker_env refuses untrusted client_instance_id "
+            f"{client_instance_id!r}: {assessment.get('reasons')}"
+        )
+    client = mwi.normalize_client_name(client_type)
+    env: dict[str, str] = {
+        PROFILE_ENV: str(profile_name),
+        CLIENT_MANAGED_ENV: "1",
+        "GITEA_MCP_CLIENT": client,
+        CLIENT_INSTANCE_ENV: assessment["client_instance_id"],
+        INSTANCE_PROVENANCE_ENV: LAUNCHER_PROVENANCE_VALUE,
+    }
+    if config_path:
+        env[CONFIG_ENV] = str(config_path)
+    if fleet_run_id:
+        env[FLEET_RUN_ENV] = str(fleet_run_id)
+    if session_id:
+        env[CLIENT_SESSION_ENV] = str(session_id)
+    if extra_env:
+        # Never let untrusted callers override the trusted instance keys.
+        protected = {
+            CLIENT_INSTANCE_ENV,
+            INSTANCE_PROVENANCE_ENV,
+            "GITEA_MCP_CLIENT",
+            CLIENT_MANAGED_ENV,
+        }
+        for key, value in extra_env.items():
+            if key in protected:
+                continue
+            env[str(key)] = str(value)
+    return env
+
+
+def resolve_launch_namespaces(
+    namespaces: "Sequence[str] | None" = None,
+) -> tuple[str, ...]:
+    """Validate an explicit project-scoped namespace subset (#985).
+
+    ``None`` keeps the historical whole-fleet behaviour and resolves to all five
+    sanctioned namespaces. An explicit sequence is validated against
+    :data:`SANCTIONED_NAMESPACES` and returned in canonical fleet order, so a
+    project-scoped configuration (for example author/reviewer/merger only) can
+    launch without inventing controller or reconciler profiles.
+
+    This is an allow-list, not a filter: anything outside the sanctioned set is
+    refused rather than silently dropped, so a typo can never quietly shrink a
+    launch to fewer workers than the operator intended.
+    """
+    if namespaces is None:
+        return tuple(SANCTIONED_NAMESPACES)
+
+    requested = [str(ns).strip().lower() for ns in namespaces]
+    if not requested or any(not ns for ns in requested):
+        raise ValueError(
+            "namespaces must be a non-empty sequence of sanctioned namespace "
+            f"names; choose from {list(SANCTIONED_NAMESPACES)}"
+        )
+
+    unknown = sorted({ns for ns in requested if ns not in SANCTIONED_NAMESPACES})
+    if unknown:
+        raise ValueError(
+            f"unsanctioned namespace(s) {unknown} requested; sanctioned "
+            f"namespaces are {list(SANCTIONED_NAMESPACES)}"
+        )
+
+    duplicates = sorted({ns for ns in requested if requested.count(ns) > 1})
+    if duplicates:
+        raise ValueError(
+            f"duplicate namespace(s) {duplicates} requested; each sanctioned "
+            "namespace may appear at most once in one launch"
+        )
+
+    selected = set(requested)
+    return tuple(ns for ns in SANCTIONED_NAMESPACES if ns in selected)
+
+
+def build_application_mcp_servers(
+    profile_by_namespace: Mapping[str, str],
+    *,
+    client_type: str | None,
+    namespaces: "Sequence[str] | None" = None,
+    config_path: str | None = None,
+    client_instance_id: str | None = None,
+    fleet_run_id: str | None = None,
+    session_id: str | None = None,
+    launch_nonce: str | None = None,
+    command: str | None = None,
+    args: list[str] | None = None,
+    now=None,
+) -> dict[str, Any]:
+    """Build a production ``mcpServers`` map for one application launch.
+
+    Mints one ``client_instance_id`` when *client_instance_id* is omitted (fresh
+    launch). When the caller supplies a previously minted trusted ID (resume of
+    the same launch / config rewrite), that ID is reused so reconnect keeps
+    attribution. A full new application restart omits the ID and receives a
+    fresh mint.
+
+    Every namespace server entry receives the **same** instance ID. Separate
+    calls with no supplied ID receive distinct IDs.
+
+    *namespaces* selects an explicitly validated project-scoped subset (#985).
+    Omitting it preserves the original whole-fleet behaviour, so existing
+    five-namespace callers are unaffected. Only the resolved namespaces need a
+    profile, and only they are started; excluded namespaces are reported so a
+    caller can prove a controller/reconciler worker was never launched.
+    """
+    resolved_namespaces = resolve_launch_namespaces(namespaces)
+
+    missing = [
+        ns for ns in resolved_namespaces if not profile_by_namespace.get(ns)
+    ]
+    if missing:
+        raise ValueError(
+            "build_application_mcp_servers requires a profile for every "
+            f"requested namespace; missing: {missing}"
+        )
+
+    if client_instance_id is None:
+        launch = mint_application_launch(
+            client_type,
+            launch_nonce=launch_nonce,
+            fleet_run_id=fleet_run_id,
+            session_id=session_id,
+            now=now,
+        )
+    else:
+        assessment = fleet.assess_instance_identity(client_instance_id)
+        if not assessment["trusted"]:
+            raise ValueError(
+                "refusing to propagate untrusted client_instance_id "
+                f"{client_instance_id!r} into production launch envs: "
+                f"{assessment.get('reasons')}"
+            )
+        launch = {
+            "client_type": mwi.normalize_client_name(client_type),
+            "client_instance_id": assessment["client_instance_id"],
+            "instance_id_provenance": LAUNCHER_PROVENANCE_VALUE,
+            "instance_identity_trusted": True,
+            "fleet_run_id": (fleet_run_id or "").strip() or None,
+            "session_id": (session_id or "").strip()
+            or f"launch-{secrets.token_hex(12)}",
+            "namespaces": list(SANCTIONED_NAMESPACES),
+            "namespace_server_names": list(NAMESPACE_SERVER_NAMES),
+        }
+
+    # The launch record describes only the namespaces this launch actually
+    # starts, so downstream attribution never implies an unstarted worker.
+    launch["namespaces"] = list(resolved_namespaces)
+    launch["namespace_server_names"] = [
+        f"gitea-{ns}" for ns in resolved_namespaces
+    ]
+
+    # Resolve command/args from the production server entry when not provided.
+    if command is None or args is None:
+        import gitea_config
+
+        cmd, cmd_args = gitea_config.server_command()
+        command = command or cmd
+        args = args if args is not None else list(cmd_args)
+
+    servers: dict[str, Any] = {}
+    shared_id = launch["client_instance_id"]
+    for namespace in resolved_namespaces:
+        server_name = f"gitea-{namespace}"
+        profile = profile_by_namespace[namespace]
+        env = namespace_worker_env(
+            profile_name=profile,
+            client_type=launch["client_type"],
+            client_instance_id=shared_id,
+            config_path=config_path,
+            fleet_run_id=launch.get("fleet_run_id"),
+            session_id=launch.get("session_id"),
+        )
+        servers[server_name] = {
+            "command": command,
+            "args": list(args),
+            "env": env,
+        }
+
+    excluded = [ns for ns in SANCTIONED_NAMESPACES if ns not in resolved_namespaces]
+    return {
+        "mcpServers": servers,
+        "launch": launch,
+        "client_instance_id": shared_id,
+        "client_type": launch["client_type"],
+        "namespaces": list(resolved_namespaces),
+        "excluded_namespaces": excluded,
+        "project_scoped": bool(excluded),
+        "shared_instance_id_across_namespaces": True,
+        "namespace_count": len(resolved_namespaces),
+    }
+
+
+def launcher_entry_for_profile(
+    profile_name: str,
+    *,
+    client_type: str | None = None,
+    config_path: str | None = None,
+    client_instance_id: str | None = None,
+    fleet_run_id: str | None = None,
+    session_id: str | None = None,
+    server_key: str = "gitea-tools",
+    launch_nonce: str | None = None,
+    now=None,
+) -> dict[str, Any]:
+    """Thin single-server production launcher entry with trusted instance ID.
+
+    Used when only one namespace is being configured. Still mints (or reuses)
+    a trusted ``client_instance_id`` so production never relies on the legacy
+    placeholder identity for normal launches.
+    """
+    import gitea_config
+
+    if client_instance_id is None:
+        launch = mint_application_launch(
+            client_type or "unknown",
+            launch_nonce=launch_nonce,
+            fleet_run_id=fleet_run_id,
+            session_id=session_id,
+            now=now,
+        )
+        client_instance_id = launch["client_instance_id"]
+        client = launch["client_type"]
+        fleet_run = launch.get("fleet_run_id")
+        session = launch.get("session_id")
+    else:
+        assessment = fleet.assess_instance_identity(client_instance_id)
+        if not assessment["trusted"]:
+            raise ValueError(
+                f"untrusted client_instance_id {client_instance_id!r}"
+            )
+        client = mwi.normalize_client_name(client_type)
+        fleet_run = (fleet_run_id or "").strip() or None
+        session = (session_id or "").strip() or None
+        client_instance_id = assessment["client_instance_id"]
+
+    command, args = gitea_config.server_command()
+    env = namespace_worker_env(
+        profile_name=profile_name,
+        client_type=client,
+        client_instance_id=client_instance_id,
+        config_path=config_path or gitea_config.DEFAULT_CONFIG_PATH,
+        fleet_run_id=fleet_run,
+        session_id=session,
+    )
+    return {
+        server_key: {
+            "command": command,
+            "args": args,
+            "env": env,
+        },
+        "client_instance_id": client_instance_id,
+        "client_type": client,
+    }
+
+
+def collect_instance_ids_from_mcp_servers(
+    mcp_servers: Mapping[str, Any],
+    *,
+    namespaces: "Sequence[str] | None" = None,
+) -> dict[str, Any]:
+    """Inspect a production mcpServers map for shared instance attribution.
+
+    Returns the unique set of client_instance_id values across the gitea-*
+    servers of one launch and whether every worker of that launch shares
+    exactly one trusted ID.
+
+    The inspected set is the launch's own namespaces, not an assumed five
+    (#985): a project-scoped author/reviewer/merger launch must be able to
+    prove shared attribution without a controller or reconciler entry counting
+    as a missing worker. Pass *namespaces* to assert an exact expected subset;
+    omit it to inspect whichever sanctioned namespace servers are present.
+    """
+    if namespaces is None:
+        expected_servers = [
+            name for name in NAMESPACE_SERVER_NAMES if name in mcp_servers
+        ]
+    else:
+        expected_servers = [
+            f"gitea-{ns}" for ns in resolve_launch_namespaces(namespaces)
+        ]
+
+    ids: list[str] = []
+    by_server: dict[str, str | None] = {}
+    for name in expected_servers:
+        entry = mcp_servers.get(name) or {}
+        env = entry.get("env") or {}
+        raw = (env.get(CLIENT_INSTANCE_ENV) or "").strip() or None
+        by_server[name] = raw
+        if raw:
+            ids.append(raw)
+    unique = sorted(set(ids))
+    trusted = [
+        i
+        for i in unique
+        if fleet.assess_instance_identity(i)["trusted"]
+    ]
+    missing_servers = [n for n in expected_servers if n not in mcp_servers]
+    return {
+        "server_instance_ids": by_server,
+        "unique_instance_ids": unique,
+        "trusted_instance_ids": trusted,
+        "inspected_servers": list(expected_servers),
+        "missing_servers": missing_servers,
+        "shared_single_trusted_id": (
+            bool(expected_servers)
+            and not missing_servers
+            and len(unique) == 1
+            and len(trusted) == 1
+            and all(by_server.get(n) == unique[0] for n in expected_servers)
+        ),
+        "namespace_server_count": sum(
+            1 for n in NAMESPACE_SERVER_NAMES if n in mcp_servers
+        ),
+    }
+
+
+def inherit_or_refuse_client_instance(
+    env: Mapping[str, str] | None = None,
+) -> dict[str, Any]:
+    """Resolve instance identity for a worker process at serve time.
+
+    Production workers inherit the launcher-issued ID. They never mint a
+    trusted ID themselves. Missing / legacy / malformed values fail soft into
+    an untrusted assessment so registration can still record a diagnostic row
+    without authorizing multi-instance fleet mutation.
+    """
+    source = dict(env if env is not None else os.environ)
+    raw = (source.get(CLIENT_INSTANCE_ENV) or "").strip() or None
+    provenance_marker = (source.get(INSTANCE_PROVENANCE_ENV) or "").strip()
+    assessment = dict(fleet.assess_instance_identity(raw))
+    # The ``inst-…`` format is public and reproducible, so format alone cannot
+    # establish trust: anyone able to set one env var could otherwise hand-write
+    # a well-formed ID and be believed. Trust therefore requires the launcher's
+    # own seal as well, and a well-formed but unsealed identity fails closed
+    # rather than degrading to a mere annotation (#985).
+    if assessment["trusted"] and provenance_marker != LAUNCHER_PROVENANCE_VALUE:
+        assessment["launcher_sealed"] = False
+        assessment["trusted"] = False
+        assessment["complete"] = False
+        assessment["provenance"] = fleet.INSTANCE_ID_PROVENANCE_UNSEALED
+        assessment["reasons"] = list(assessment.get("reasons") or []) + [
+            f"{INSTANCE_PROVENANCE_ENV} is not {LAUNCHER_PROVENANCE_VALUE!r}; "
+            "identity format is valid but not sealed by the production "
+            "launcher, so trusted attribution is refused"
+        ]
+    else:
+        assessment["launcher_sealed"] = bool(
+            assessment["trusted"]
+            and provenance_marker == LAUNCHER_PROVENANCE_VALUE
+        )
+    assessment["fleet_run_id"] = (source.get(FLEET_RUN_ENV) or "").strip() or None
+    assessment["session_id"] = (
+        (source.get(CLIENT_SESSION_ENV) or "").strip() or None
+    )
+    assessment["client_type"] = mwi.normalize_client_name(
+        (source.get("GITEA_MCP_CLIENT") or "").strip() or None
+    )
+    return assessment
+
+
+# ---------------------------------------------------------------------------
+# Runnable launch path (#985)
+# ---------------------------------------------------------------------------
+
+#: How each supported LLM client is started with a per-launch MCP config.
+#:
+#: The launch interface is deliberately a data-driven registry rather than a
+#: Claude-specific code path: adding another supported client is one entry, and
+#: every client necessarily goes through the same mint-once/propagate-to-all
+#: identity mechanism because the argv builder only ever receives a config file
+#: this module wrote.
+CLIENT_LAUNCH_SPECS: dict[str, dict[str, Any]] = {
+    "claude_code": {
+        "command": "claude",
+        "config_args": lambda path: ["--mcp-config", path, "--strict-mcp-config"],
+        "docs": "claude --mcp-config <per-launch.json> --strict-mcp-config",
+    },
+}
+
+
+def supported_launch_clients() -> tuple[str, ...]:
+    """Client types with a sanctioned runnable launch path."""
+    return tuple(sorted(CLIENT_LAUNCH_SPECS))
+
+
+def build_client_launch_argv(
+    client_type: str,
+    config_path: str,
+    *,
+    extra_args: "Sequence[str] | None" = None,
+) -> list[str]:
+    """Build the argv that starts *client_type* against a per-launch config."""
+    client = mwi.normalize_client_name(client_type)
+    spec = CLIENT_LAUNCH_SPECS.get(client)
+    if spec is None:
+        raise ValueError(
+            f"no sanctioned runnable launch path for client {client!r}; "
+            f"supported clients: {list(supported_launch_clients())}"
+        )
+    argv = [str(spec["command"])] + list(spec["config_args"](str(config_path)))
+    if extra_args:
+        argv.extend(str(a) for a in extra_args)
+    return argv
+
+
+def write_launch_config(
+    mcp_servers: Mapping[str, Any],
+    *,
+    directory: str | None = None,
+) -> str:
+    """Write one launch's ``mcpServers`` config to a fresh per-launch file.
+
+    The file is per-launch and owner-readable only. #985 explicitly rules out
+    persisting an instance ID into a shared, reused ``.mcp.json``: two
+    concurrent sessions reading one static trusted ID is precisely the reuse
+    case the duplicate guard must reject. A distinct file per launch keeps the
+    minted identity bound to the launch that minted it.
+    """
+    import json
+    import tempfile
+
+    fd, path = tempfile.mkstemp(
+        prefix="gitea-mcp-launch-", suffix=".json", dir=directory
+    )
+    try:
+        os.fchmod(fd, 0o600)
+        with os.fdopen(fd, "w", encoding="utf-8") as handle:
+            json.dump({"mcpServers": dict(mcp_servers)}, handle, indent=2)
+            handle.write("\n")
+    except Exception:
+        try:
+            os.unlink(path)
+        except OSError:
+            pass
+        raise
+    return path
+
+
+def prepare_application_launch(
+    profile_by_namespace: Mapping[str, str],
+    *,
+    client_type: str,
+    namespaces: "Sequence[str] | None" = None,
+    config_path: str | None = None,
+    fleet_run_id: str | None = None,
+    session_id: str | None = None,
+    launch_nonce: str | None = None,
+    extra_args: "Sequence[str] | None" = None,
+    config_directory: str | None = None,
+    now=None,
+) -> dict[str, Any]:
+    """Mint one launch and produce everything needed to exec the client.
+
+    Returns the built servers, the per-launch config file path, the argv to
+    exec, and the shared trusted identity — without starting anything, so the
+    same preparation is testable and inspectable via ``--dry-run``.
+    """
+    built = build_application_mcp_servers(
+        profile_by_namespace,
+        client_type=client_type,
+        namespaces=namespaces,
+        config_path=config_path,
+        fleet_run_id=fleet_run_id,
+        session_id=session_id,
+        launch_nonce=launch_nonce,
+        now=now,
+    )
+    # Fail before writing anything if this client has no runnable path.
+    argv_probe = build_client_launch_argv(
+        client_type, "<pending>", extra_args=extra_args
+    )
+    launch_config_path = write_launch_config(
+        built["mcpServers"], directory=config_directory
+    )
+    argv = build_client_launch_argv(
+        client_type, launch_config_path, extra_args=extra_args
+    )
+    proof = collect_instance_ids_from_mcp_servers(
+        built["mcpServers"], namespaces=built["namespaces"]
+    )
+    return {
+        "client_type": built["client_type"],
+        "client_instance_id": built["client_instance_id"],
+        "namespaces": built["namespaces"],
+        "excluded_namespaces": built["excluded_namespaces"],
+        "project_scoped": built["project_scoped"],
+        "launch_config_path": launch_config_path,
+        "argv": argv,
+        "argv_template": argv_probe,
+        "mcpServers": built["mcpServers"],
+        "launch": built["launch"],
+        "shared_single_trusted_id": proof["shared_single_trusted_id"],
+        "attribution_proof": proof,
+    }
+
+
+def _parse_profile_assignments(values: "Sequence[str]") -> dict[str, str]:
+    """Parse ``namespace=profile`` CLI pairs into a mapping."""
+    profiles: dict[str, str] = {}
+    for raw in values or ():
+        text = str(raw)
+        if "=" not in text:
+            raise ValueError(
+                f"invalid --profile {text!r}; expected namespace=profile-name"
+            )
+        namespace, _, profile = text.partition("=")
+        namespace = namespace.strip().lower()
+        profile = profile.strip()
+        if not namespace or not profile:
+            raise ValueError(
+                f"invalid --profile {text!r}; expected namespace=profile-name"
+            )
+        if namespace in profiles:
+            raise ValueError(f"duplicate --profile entry for namespace {namespace!r}")
+        profiles[namespace] = profile
+    return profiles
+
+
+def main(argv: "Sequence[str] | None" = None) -> int:
+    """Runnable entry point: start a supported LLM client for one launch.
+
+    Example (project-scoped, three namespaces)::
+
+        python3 -m mcp_application_launcher \\
+            --client claude_code \\
+            --namespaces author,reviewer,merger \\
+            --profile author=prgs-author \\
+            --profile reviewer=prgs-reviewer \\
+            --profile merger=prgs-merger
+    """
+    import argparse
+    import json
+
+    parser = argparse.ArgumentParser(
+        prog="mcp_application_launcher",
+        description=(
+            "Start a supported LLM client with one trusted, per-launch "
+            "GITEA_MCP_CLIENT_INSTANCE shared by every MCP namespace worker."
+        ),
+    )
+    parser.add_argument(
+        "--client",
+        default="claude_code",
+        help=f"client to launch; supported: {list(supported_launch_clients())}",
+    )
+    parser.add_argument(
+        "--namespaces",
+        default=None,
+        help=(
+            "comma-separated sanctioned namespace subset (e.g. "
+            "'author,reviewer,merger'); omit to launch all five"
+        ),
+    )
+    parser.add_argument(
+        "--profile",
+        action="append",
+        default=[],
+        metavar="NAMESPACE=PROFILE",
+        help="profile for one namespace; repeat once per launched namespace",
+    )
+    parser.add_argument("--config-path", default=None, help="profiles.json path")
+    parser.add_argument("--fleet-run-id", default=None)
+    parser.add_argument("--session-id", default=None)
+    parser.add_argument(
+        "--dry-run",
+        action="store_true",
+        help="print the launch plan as JSON and exit without starting anything",
+    )
+    parser.add_argument(
+        "client_args",
+        nargs=argparse.REMAINDER,
+        help="arguments forwarded to the client after '--'",
+    )
+    ns = parser.parse_args(list(argv) if argv is not None else None)
+
+    try:
+        profiles = _parse_profile_assignments(ns.profile)
+        requested = (
+            [part.strip() for part in ns.namespaces.split(",") if part.strip()]
+            if ns.namespaces
+            else None
+        )
+        resolved = resolve_launch_namespaces(requested)
+        # Profiles for namespaces that are not being launched are almost always
+        # a mistake (a stale five-namespace invocation), so refuse rather than
+        # silently ignore them.
+        stray = sorted(set(profiles) - set(resolved))
+        if stray:
+            raise ValueError(
+                f"--profile given for namespace(s) {stray} that this launch "
+                f"does not start; launched namespaces are {list(resolved)}"
+            )
+        forwarded = [a for a in (ns.client_args or []) if a != "--"]
+        prepared = prepare_application_launch(
+            profiles,
+            client_type=ns.client,
+            namespaces=requested,
+            config_path=ns.config_path,
+            fleet_run_id=ns.fleet_run_id,
+            session_id=ns.session_id,
+            extra_args=forwarded,
+        )
+    except ValueError as exc:
+        parser.error(str(exc))
+        return 2  # pragma: no cover - argparse.error raises SystemExit
+
+    if ns.dry_run:
+        plan = {
+            key: prepared[key]
+            for key in (
+                "client_type",
+                "client_instance_id",
+                "namespaces",
+                "excluded_namespaces",
+                "project_scoped",
+                "launch_config_path",
+                "argv",
+                "shared_single_trusted_id",
+            )
+        }
+        print(json.dumps(plan, indent=2))
+        return 0
+
+    argv_to_exec = prepared["argv"]
+    os.execvp(argv_to_exec[0], argv_to_exec)
+    return 0  # pragma: no cover - execvp does not return
+
+
+if __name__ == "__main__":  # pragma: no cover - process entry point
+    raise SystemExit(main())

mcp_fleet_snapshot.py

@@ -0,0 +1,873 @@
+"""Instance-level fleet identity and health snapshots (#978).
+
+#948 established per-worker ownership; #975 made heartbeats keep those rows
+live. Neither surface could enumerate the fleet at *instance* granularity:
+which application launch owns which five namespace workers, whether two
+Codex launches are distinct, or whether a live collision is real rather than
+a shared client type.
+
+This module is pure. Callers supply registry rows (and optional enrichments);
+nothing here opens SQLite, scans process tables, or mutates state. Production
+evidence for the fleet gate is the snapshot returned by the sanctioned
+controller/reconciler tool that wraps this assessor.
+
+Identity hierarchy (highest → lowest):
+
+* ``client_type`` — application family (``codex``, ``claude_code``, …)
+* ``client_instance_id`` — one running application launch (trusted launcher)
+* ``fleet_run_id`` — operator-approved enrollment / canary cohort
+* ``worker_id`` / ``worker_identity`` — one namespace worker process
+* ``namespace`` — author | reviewer | merger | controller | reconciler
+
+Multiple simultaneous instances of the same ``client_type`` are first-class.
+Sharing only a profile or client type is never a duplicate.
+"""
+
+from __future__ import annotations
+
+import hashlib
+import secrets
+from collections import defaultdict
+from datetime import datetime, timezone
+from typing import Any, Callable, Iterable, Mapping
+
+import mcp_worker_identity as mwi
+
+# --- Classification labels ------------------------------------------------
+
+CLASS_EXPECTED = "expected_enrolled"
+CLASS_MISSING = "missing_expected"
+CLASS_UNMANIFESTED = "unmanifested"
+CLASS_DUPLICATE_NAMESPACE = "duplicate_namespace_worker"
+CLASS_INSTANCE_ID_COLLISION = "instance_id_collision"
+CLASS_WORKER_ID_COLLISION = "worker_identity_collision"
+CLASS_SESSION_COLLISION = "session_identity_collision"
+CLASS_GENERATION_COLLISION = "generation_identity_collision"
+CLASS_PROCESS_COLLISION = "process_identity_collision"
+CLASS_PID_COLLISION = "pid_collision"
+CLASS_OWNERSHIP_COLLISION = "ownership_fencing_collision"
+CLASS_ORPHANED = "orphaned_unowned"
+CLASS_UNKNOWN_CLIENT = "unknown_client"
+CLASS_FOREIGN_REPOSITORY = "foreign_repository"
+CLASS_OLD_REVISION = "old_revision"
+CLASS_STALE_WORKER = "stale_orphaned_worker"
+CLASS_LEGACY_INCOMPLETE = "legacy_incomplete_identity"
+CLASS_HISTORICAL = "historical_dead"
+CLASS_HEALTHY = "healthy"
+
+#: Active blockers that make the live fleet unsafe for mutation-gated work.
+ACTIVE_BLOCKER_CLASSES = frozenset(
+    {
+        CLASS_MISSING,
+        CLASS_UNMANIFESTED,
+        CLASS_DUPLICATE_NAMESPACE,
+        CLASS_INSTANCE_ID_COLLISION,
+        CLASS_WORKER_ID_COLLISION,
+        CLASS_SESSION_COLLISION,
+        CLASS_GENERATION_COLLISION,
+        CLASS_PROCESS_COLLISION,
+        CLASS_PID_COLLISION,
+        CLASS_OWNERSHIP_COLLISION,
+        CLASS_ORPHANED,
+        CLASS_UNKNOWN_CLIENT,
+        CLASS_FOREIGN_REPOSITORY,
+        CLASS_OLD_REVISION,
+        CLASS_STALE_WORKER,
+        CLASS_LEGACY_INCOMPLETE,
+    }
+)
+
+SANCTIONED_NAMESPACES = frozenset(
+    {"author", "reviewer", "merger", "controller", "reconciler"}
+)
+
+INSTANCE_ID_PROVENANCE_TRUSTED = "trusted_launcher"
+INSTANCE_ID_PROVENANCE_LEGACY = "legacy_incomplete"
+INSTANCE_ID_PROVENANCE_MISSING = "missing"
+#: Well-formed ``inst-…`` identity presented without the launcher's own
+#: provenance seal (#985). The format alone is public and reproducible, so an
+#: unsealed value is treated as manually asserted trust and fails closed.
+INSTANCE_ID_PROVENANCE_UNSEALED = "unsealed_launcher"
+
+CLIENT_INSTANCE_ENV = "GITEA_MCP_CLIENT_INSTANCE"
+FLEET_RUN_ENV = "GITEA_MCP_FLEET_RUN_ID"
+PROCESS_IDENTITY_ENV = "GITEA_MCP_PROCESS_IDENTITY"
+INSTANCE_PROVENANCE_ENV = "GITEA_MCP_INSTANCE_PROVENANCE"
+
+_LEGACY_INSTANCE_PREFIXES = ("pid-", "proc-", "legacy-")
+#: Trusted launcher-issued IDs use the reserved ``inst-`` prefix
+#: (see :func:`generate_client_instance_id`). Ordinary user-supplied strings
+#: without that prefix never count as trusted attribution.
+_TRUSTED_INSTANCE_PREFIX = "inst-"
+
+
+def _utc_now() -> datetime:
+    return datetime.now(timezone.utc)
+
+
+def _ts(value: datetime) -> str:
+    return value.astimezone(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ")
+
+
+def generate_client_instance_id(
+    client_type: str | None,
+    *,
+    launch_nonce: str | None = None,
+    now: datetime | None = None,
+) -> str:
+    """Mint a distinct instance ID for one application launch (#978).
+
+    The trusted launcher (or host that starts all five namespace workers)
+    generates this once per launch and injects it as ``GITEA_MCP_CLIENT_INSTANCE``
+    into every worker environment. Workers never invent their own instance ID
+    from PID proximity or timestamps.
+    """
+    client = mwi.normalize_client_name(client_type)
+    stamp = (now or _utc_now()).astimezone(timezone.utc).strftime(
+        mwi.IDENTITY_TIMESTAMP_FORMAT
+    )
+    nonce = launch_nonce if launch_nonce is not None else secrets.token_hex(16)
+    digest = hashlib.sha256(
+        f"{client}\x1f{stamp}\x1f{nonce}".encode("utf-8")
+    ).hexdigest()[:12]
+    return f"inst-{client}-{stamp}-{digest}"
+
+
+def assess_instance_identity(
+    raw_instance_id: str | None,
+    *,
+    source: str | None = None,
+) -> dict[str, Any]:
+    """Classify whether a client_instance_id is trusted enough for mutation.
+
+    Trusted instance IDs are non-empty, match the launcher-minted ``inst-…``
+    format from :func:`generate_client_instance_id`, and are not pre-#978
+    PID/proc/legacy placeholders. Ordinary user-supplied or malformed values
+    fail closed as untrusted so they cannot spoof multi-instance attribution.
+    Incomplete identities remain visible for diagnosis but cannot authorize
+    unsafe mutation.
+    """
+    text = (raw_instance_id or "").strip()
+    if not text:
+        return {
+            "client_instance_id": None,
+            "complete": False,
+            "trusted": False,
+            "provenance": INSTANCE_ID_PROVENANCE_MISSING,
+            "reasons": [
+                "client_instance_id is missing; the trusted launcher must set "
+                f"{CLIENT_INSTANCE_ENV} once per application launch"
+            ],
+        }
+    lowered = text.lower()
+    if lowered.startswith(_LEGACY_INSTANCE_PREFIXES) or source == "pid_fallback":
+        return {
+            "client_instance_id": text,
+            "complete": False,
+            "trusted": False,
+            "provenance": INSTANCE_ID_PROVENANCE_LEGACY,
+            "reasons": [
+                f"client_instance_id {text!r} is a legacy PID/process fallback, "
+                "not a trusted launcher-issued instance identity"
+            ],
+        }
+    if not text.startswith(_TRUSTED_INSTANCE_PREFIX) or len(text) <= len(
+        _TRUSTED_INSTANCE_PREFIX
+    ):
+        return {
+            "client_instance_id": text,
+            "complete": False,
+            "trusted": False,
+            "provenance": INSTANCE_ID_PROVENANCE_LEGACY,
+            "reasons": [
+                f"client_instance_id {text!r} is malformed or user-supplied and "
+                f"does not use the trusted launcher prefix "
+                f"{_TRUSTED_INSTANCE_PREFIX!r}; refusing trusted attribution"
+            ],
+        }
+    return {
+        "client_instance_id": text,
+        "complete": True,
+        "trusted": True,
+        "provenance": INSTANCE_ID_PROVENANCE_TRUSTED,
+        "reasons": [],
+    }
+
+
+def resolve_client_instance_from_env(
+    env: Mapping[str, str] | None = None,
+    *,
+    pid: int | None = None,
+) -> dict[str, Any]:
+    """Resolve instance identity from launcher env without inventing one.
+
+    When the trusted key is absent, return incomplete evidence rather than a
+    silent ``pid-<n>`` identity. Callers that still need a non-empty registry
+    key may choose a legacy placeholder deliberately; they must not treat it as
+    trusted.
+    """
+    source = dict(env or {})
+    raw = (source.get(CLIENT_INSTANCE_ENV) or "").strip()
+    assessment = assess_instance_identity(raw or None)
+    assessment["fleet_run_id"] = (source.get(FLEET_RUN_ENV) or "").strip() or None
+    assessment["process_identity"] = (
+        (source.get(PROCESS_IDENTITY_ENV) or "").strip()
+        or (f"pid-{pid}" if pid is not None else None)
+    )
+    return assessment
+
+
+def _public_worker(record: Mapping[str, Any]) -> dict[str, Any]:
+    return {
+        "worker_id": record.get("worker_identity") or record.get("worker_id"),
+        "worker_identity": record.get("worker_identity") or record.get("worker_id"),
+        "client_type": mwi.normalize_client_name(
+            record.get("client_name") or record.get("client_type")
+        ),
+        "client_instance_id": record.get("client_instance_id"),
+        "fleet_run_id": record.get("fleet_run_id"),
+        "namespace": record.get("namespace"),
+        "profile": record.get("profile"),
+        "declared_role": record.get("role") or record.get("declared_role"),
+        "authenticated_account": record.get("authenticated_account"),
+        "session_id": record.get("session_id"),
+        "generation_id": record.get("generation_id"),
+        "process_identity": record.get("process_identity")
+        or (
+            f"pid-{record['pid']}"
+            if record.get("pid") is not None
+            else None
+        ),
+        "pid": record.get("pid"),
+        "repository_binding": record.get("repository_binding"),
+        "remote": record.get("remote"),
+        "startup_revision": record.get("startup_revision"),
+        "loaded_revision": record.get("loaded_revision"),
+        "parity_revision": record.get("parity_revision"),
+        "live_revision": record.get("live_revision"),
+        "runtime_provenance": record.get("runtime_provenance")
+        or record.get("transport"),
+        "transport": record.get("transport"),
+        "started_at": record.get("started_at"),
+        "last_heartbeat_at": record.get("last_heartbeat_at"),
+        "heartbeat_ttl_seconds": record.get("heartbeat_ttl_seconds"),
+        "fencing_epoch": record.get("fencing_epoch"),
+        "status": record.get("status"),
+        "instance_id_provenance": record.get("instance_id_provenance"),
+    }
+
+
+def _liveness(
+    record: Mapping[str, Any],
+    *,
+    now: datetime | None,
+    pid_alive_probe: Callable[[int | None], bool | None] | None,
+) -> dict[str, Any]:
+    pid_alive = None
+    if pid_alive_probe is not None and record.get("pid") is not None:
+        try:
+            pid_alive = pid_alive_probe(record.get("pid"))
+        except Exception:
+            pid_alive = None
+    return mwi.WorkerRegistry.is_live(record, now=now, pid_alive=pid_alive)
+
+
+def _consistency_token(rows: Iterable[Mapping[str, Any]], snapshot_at: str) -> str:
+    material = [snapshot_at]
+    for row in sorted(
+        rows,
+        key=lambda r: (
+            str(r.get("worker_identity") or ""),
+            str(r.get("last_heartbeat_at") or ""),
+            str(r.get("fencing_epoch") or ""),
+        ),
+    ):
+        material.append(
+            "|".join(
+                [
+                    str(row.get("worker_identity") or ""),
+                    str(row.get("client_instance_id") or ""),
+                    str(row.get("status") or ""),
+                    str(row.get("last_heartbeat_at") or ""),
+                    str(row.get("fencing_epoch") or ""),
+                    str(row.get("generation_id") or ""),
+                ]
+            )
+        )
+    digest = hashlib.sha256("\n".join(material).encode("utf-8")).hexdigest()[:16]
+    return f"fleetrev-{digest}"
+
+
+def build_worker_snapshot_row(
+    record: Mapping[str, Any],
+    *,
+    now: datetime | None = None,
+    pid_alive_probe: Callable[[int | None], bool | None] | None = None,
+    canonical_repository: str | None = None,
+    expected_live_revision: str | None = None,
+    heartbeat_supervised: bool | None = None,
+) -> dict[str, Any]:
+    """One point-in-time worker row for the fleet snapshot."""
+    stamp = now or _utc_now()
+    base = _public_worker(record)
+    identity = assess_instance_identity(
+        base.get("client_instance_id"),
+        source=record.get("instance_id_source"),
+    )
+    liveness = _liveness(record, now=stamp, pid_alive_probe=pid_alive_probe)
+    is_historical = str(record.get("status") or "") != mwi.STATUS_ACTIVE
+    live = bool(liveness.get("live")) and not is_historical
+
+    repo = (base.get("repository_binding") or "").strip() or None
+    foreign_repo = bool(
+        canonical_repository
+        and repo
+        and repo.rstrip("/") != str(canonical_repository).rstrip("/")
+    )
+    old_revision = False
+    if expected_live_revision:
+        for key in ("startup_revision", "loaded_revision", "parity_revision", "live_revision"):
+            rev = (base.get(key) or "").strip()
+            if rev and rev != expected_live_revision:
+                old_revision = True
+                break
+
+    ownership_state = "historical" if is_historical else (
+        "live" if live else "stale"
+    )
+    if live and not identity["trusted"]:
+        ownership_state = "live_untrusted_identity"
+    if live and not base.get("session_id"):
+        ownership_state = "orphaned"
+
+    mutation_safe = bool(
+        live
+        and identity["trusted"]
+        and not foreign_repo
+        and not old_revision
+        and ownership_state == "live"
+        and base.get("client_type") != mwi.UNKNOWN_CLIENT
+    )
+
+    restart_required = bool(
+        old_revision
+        or (live and not liveness.get("heartbeat_fresh", True))
+    )
+
+    return {
+        **base,
+        "instance_identity": identity,
+        "client_instance_id": identity["client_instance_id"] or base.get("client_instance_id"),
+        "instance_id_provenance": identity["provenance"],
+        "instance_identity_trusted": identity["trusted"],
+        "live": live,
+        "historical": is_historical,
+        "liveness": liveness,
+        "heartbeat": {
+            "registered": bool(base.get("last_heartbeat_at")),
+            "supervised": heartbeat_supervised,
+            "age_seconds": liveness.get("heartbeat_age_seconds"),
+            "ttl_seconds": liveness.get("heartbeat_ttl_seconds"),
+            "fresh": liveness.get("heartbeat_fresh"),
+            "last_heartbeat_at": base.get("last_heartbeat_at"),
+        },
+        "fencing": {
+            "fencing_epoch": base.get("fencing_epoch"),
+            "generation_id": base.get("generation_id"),
+        },
+        "ownership_state": ownership_state,
+        "foreign_repository": foreign_repo,
+        "old_revision": old_revision,
+        "stale": not live and not is_historical,
+        "restart_required": restart_required,
+        "mutation_safe": mutation_safe,
+        "conflicting_live_sessions": [],
+    }
+
+
+def _collision_groups(
+    live_rows: list[dict[str, Any]],
+    key_fn,
+) -> dict[str, list[dict[str, Any]]]:
+    groups: dict[str, list[dict[str, Any]]] = defaultdict(list)
+    for row in live_rows:
+        key = key_fn(row)
+        if key is None or key == "" or key == "None":
+            continue
+        groups[str(key)].append(row)
+    return {k: v for k, v in groups.items() if len(v) > 1}
+
+
+def snapshot_instance_fleet(
+    records: Iterable[Mapping[str, Any]],
+    *,
+    expected_manifest: list[Mapping[str, Any]] | None = None,
+    now: datetime | None = None,
+    pid_alive_probe: Callable[[int | None], bool | None] | None = None,
+    canonical_repository: str | None = None,
+    expected_live_revision: str | None = None,
+    registry_revision: str | None = None,
+    known_client_types: Iterable[str] | None = None,
+) -> dict[str, Any]:
+    """Authoritative point-in-time fleet snapshot with classification (#978).
+
+    Historical dead rows are reported separately and never automatically make
+    the live fleet unsafe.
+    """
+    stamp = now or _utc_now()
+    snapshot_at = _ts(stamp)
+    known = {
+        mwi.normalize_client_name(c)
+        for c in (known_client_types or mwi.CLIENT_ALIASES.values())
+    }
+    known.discard(mwi.UNKNOWN_CLIENT)
+
+    all_rows: list[dict[str, Any]] = []
+    for record in records:
+        all_rows.append(
+            build_worker_snapshot_row(
+                record,
+                now=stamp,
+                pid_alive_probe=pid_alive_probe,
+                canonical_repository=canonical_repository,
+                expected_live_revision=expected_live_revision,
+            )
+        )
+
+    live_rows = [r for r in all_rows if r["live"]]
+    historical_rows = [r for r in all_rows if r["historical"]]
+    stale_rows = [r for r in all_rows if r["stale"]]
+
+    # --- identity collisions among live workers ---
+    findings: list[dict[str, Any]] = []
+
+    def _finding(
+        classification: str,
+        *,
+        severity: str,
+        workers: list[dict[str, Any]] | None = None,
+        instance_ids: list[str] | None = None,
+        detail: str,
+        active_blocker: bool,
+    ) -> None:
+        findings.append(
+            {
+                "classification": classification,
+                "severity": severity,
+                "active_blocker": active_blocker,
+                "detail": detail,
+                "client_instance_ids": instance_ids or sorted(
+                    {
+                        str(w.get("client_instance_id"))
+                        for w in (workers or [])
+                        if w.get("client_instance_id")
+                    }
+                ),
+                "worker_identities": [
+                    w.get("worker_identity") for w in (workers or [])
+                ],
+            }
+        )
+
+    # Duplicate worker identity (should not happen with PK, still detect)
+    for wid, group in _collision_groups(
+        live_rows, lambda r: r.get("worker_identity")
+    ).items():
+        _finding(
+            CLASS_WORKER_ID_COLLISION,
+            severity="blocker",
+            workers=group,
+            detail=f"worker identity {wid!r} is claimed by {len(group)} live workers",
+            active_blocker=True,
+        )
+
+    # Reused session identity across live workers
+    for sid, group in _collision_groups(live_rows, lambda r: r.get("session_id")).items():
+        # Same session may appear once; collision only when multiple workers share it
+        # across different worker identities (always true for group size > 1).
+        _finding(
+            CLASS_SESSION_COLLISION,
+            severity="blocker",
+            workers=group,
+            detail=f"session identity {sid!r} is reused by {len(group)} live workers",
+            active_blocker=True,
+        )
+
+    # Generation claimed by multiple live sessions/workers is a conflict when
+    # the workers are not the five sanctioned namespaces of one instance.
+    for gen, group in _collision_groups(
+        live_rows, lambda r: r.get("generation_id")
+    ).items():
+        namespaces = {g.get("namespace") for g in group if g.get("namespace")}
+        instance_ids = {g.get("client_instance_id") for g in group}
+        # Multiple workers under one generation is only valid if they share one
+        # instance and distinct namespaces. Same generation + same namespace = bad.
+        by_ns: dict[str, list] = defaultdict(list)
+        for g in group:
+            by_ns[str(g.get("namespace") or "")].append(g)
+        ns_dups = {ns: rows for ns, rows in by_ns.items() if ns and len(rows) > 1}
+        if ns_dups or len(instance_ids) > 1:
+            _finding(
+                CLASS_GENERATION_COLLISION,
+                severity="blocker",
+                workers=group,
+                detail=(
+                    f"generation {gen!r} is contested across namespaces/instances "
+                    f"(namespaces={sorted(namespaces)}, "
+                    f"instances={sorted(str(i) for i in instance_ids if i)})"
+                ),
+                active_blocker=True,
+            )
+
+    # Process identity / PID collisions across distinct workers
+    for proc, group in _collision_groups(
+        live_rows, lambda r: r.get("process_identity")
+    ).items():
+        if len({r.get("worker_identity") for r in group}) > 1:
+            _finding(
+                CLASS_PROCESS_COLLISION,
+                severity="blocker",
+                workers=group,
+                detail=f"process identity {proc!r} is shared by distinct live workers",
+                active_blocker=True,
+            )
+
+    for pid, group in _collision_groups(live_rows, lambda r: r.get("pid")).items():
+        if len({r.get("worker_identity") for r in group}) > 1:
+            _finding(
+                CLASS_PID_COLLISION,
+                severity="blocker",
+                workers=group,
+                detail=f"PID {pid} is shared by distinct live workers",
+                active_blocker=True,
+            )
+
+    # Fencing/ownership: same fencing epoch on different workers of different instances
+    for epoch, group in _collision_groups(
+        live_rows,
+        lambda r: (
+            f"{r.get('generation_id')}:{r.get('fencing_epoch')}"
+            if r.get("generation_id") is not None and r.get("fencing_epoch") is not None
+            else None
+        ),
+    ).items():
+        if len({r.get("client_instance_id") for r in group}) > 1:
+            _finding(
+                CLASS_OWNERSHIP_COLLISION,
+                severity="blocker",
+                workers=group,
+                detail=(
+                    f"fencing token {epoch!r} spans more than one client_instance_id"
+                ),
+                active_blocker=True,
+            )
+
+    # Per-instance grouping
+    by_instance: dict[str, list[dict[str, Any]]] = defaultdict(list)
+    unkeyed_live: list[dict[str, Any]] = []
+    for row in live_rows:
+        iid = row.get("client_instance_id")
+        if not iid:
+            unkeyed_live.append(row)
+            continue
+        by_instance[str(iid)].append(row)
+
+    instances: list[dict[str, Any]] = []
+    for iid, workers in sorted(by_instance.items()):
+        client_types = sorted({w.get("client_type") for w in workers if w.get("client_type")})
+        trusted = all(w.get("instance_identity_trusted") for w in workers)
+        namespaces = [w.get("namespace") for w in workers]
+        ns_counts: dict[str, int] = defaultdict(int)
+        for ns in namespaces:
+            if ns:
+                ns_counts[str(ns)] += 1
+        dup_ns = sorted(ns for ns, n in ns_counts.items() if n > 1)
+        if dup_ns:
+            _finding(
+                CLASS_DUPLICATE_NAMESPACE,
+                severity="blocker",
+                workers=[w for w in workers if w.get("namespace") in dup_ns],
+                instance_ids=[iid],
+                detail=(
+                    f"instance {iid!r} has more than one live worker for "
+                    f"namespace(s) {dup_ns}"
+                ),
+                active_blocker=True,
+            )
+
+        # Live reuse of one instance ID with incompatible client types
+        if len(client_types) > 1:
+            _finding(
+                CLASS_INSTANCE_ID_COLLISION,
+                severity="blocker",
+                workers=workers,
+                instance_ids=[iid],
+                detail=(
+                    f"client_instance_id {iid!r} is live under multiple client "
+                    f"types {client_types}"
+                ),
+                active_blocker=True,
+            )
+
+        if not trusted:
+            _finding(
+                CLASS_LEGACY_INCOMPLETE,
+                severity="blocker",
+                workers=workers,
+                instance_ids=[iid],
+                detail=(
+                    f"instance {iid!r} lacks trusted launcher-issued instance "
+                    "identity; diagnostic reads remain available"
+                ),
+                active_blocker=True,
+            )
+
+        unknown = [w for w in workers if w.get("client_type") == mwi.UNKNOWN_CLIENT]
+        if unknown:
+            _finding(
+                CLASS_UNKNOWN_CLIENT,
+                severity="blocker",
+                workers=unknown,
+                instance_ids=[iid],
+                detail=f"instance {iid!r} has worker(s) with unknown client_type",
+                active_blocker=True,
+            )
+
+        foreign = [w for w in workers if w.get("foreign_repository")]
+        if foreign:
+            _finding(
+                CLASS_FOREIGN_REPOSITORY,
+                severity="blocker",
+                workers=foreign,
+                instance_ids=[iid],
+                detail=f"instance {iid!r} has foreign-repository workers",
+                active_blocker=True,
+            )
+
+        old = [w for w in workers if w.get("old_revision")]
+        if old:
+            _finding(
+                CLASS_OLD_REVISION,
+                severity="blocker",
+                workers=old,
+                instance_ids=[iid],
+                detail=f"instance {iid!r} has old-revision workers",
+                active_blocker=True,
+            )
+
+        orphans = [w for w in workers if w.get("ownership_state") == "orphaned"]
+        if orphans:
+            _finding(
+                CLASS_ORPHANED,
+                severity="blocker",
+                workers=orphans,
+                instance_ids=[iid],
+                detail=f"instance {iid!r} has orphaned/unowned workers",
+                active_blocker=True,
+            )
+
+        instances.append(
+            {
+                "client_instance_id": iid,
+                "client_types": client_types,
+                "client_type": client_types[0] if len(client_types) == 1 else None,
+                "fleet_run_ids": sorted(
+                    {w.get("fleet_run_id") for w in workers if w.get("fleet_run_id")}
+                ),
+                "worker_count": len(workers),
+                "namespaces": sorted({n for n in namespaces if n}),
+                "namespace_counts": dict(ns_counts),
+                "duplicate_namespaces": dup_ns,
+                "trusted_instance_identity": trusted,
+                "workers": workers,
+                "mutation_safe": all(w.get("mutation_safe") for w in workers)
+                and not dup_ns
+                and trusted,
+            }
+        )
+
+    for row in unkeyed_live:
+        _finding(
+            CLASS_LEGACY_INCOMPLETE,
+            severity="blocker",
+            workers=[row],
+            detail="live worker has no client_instance_id",
+            active_blocker=True,
+        )
+        if row.get("client_type") == mwi.UNKNOWN_CLIENT:
+            _finding(
+                CLASS_UNKNOWN_CLIENT,
+                severity="blocker",
+                workers=[row],
+                detail="live worker has unknown client_type and no instance id",
+                active_blocker=True,
+            )
+
+    for row in stale_rows:
+        _finding(
+            CLASS_STALE_WORKER,
+            severity="warning",
+            workers=[row],
+            detail=(
+                f"worker {row.get('worker_identity')!r} is active in the registry "
+                "but not live (stale heartbeat or dead pid)"
+            ),
+            active_blocker=True,
+        )
+
+    for row in historical_rows:
+        _finding(
+            CLASS_HISTORICAL,
+            severity="info",
+            workers=[row],
+            detail=(
+                f"historical registration {row.get('worker_identity')!r} "
+                f"(status={row.get('status')!r}) is not an active blocker"
+            ),
+            active_blocker=False,
+        )
+
+    # Manifest comparison
+    expected = list(expected_manifest or [])
+    expected_ids = {
+        str(item.get("client_instance_id")).strip()
+        for item in expected
+        if (item.get("client_instance_id") or "").strip()
+    }
+    live_ids = set(by_instance.keys())
+    missing_ids = sorted(expected_ids - live_ids)
+    unmanifested_ids = sorted(live_ids - expected_ids) if expected_ids else []
+
+    for iid in missing_ids:
+        _finding(
+            CLASS_MISSING,
+            severity="blocker",
+            instance_ids=[iid],
+            detail=f"expected enrolled instance {iid!r} is missing from the live fleet",
+            active_blocker=True,
+        )
+    for iid in unmanifested_ids:
+        _finding(
+            CLASS_UNMANIFESTED,
+            severity="blocker",
+            instance_ids=[iid],
+            workers=by_instance.get(iid, []),
+            detail=(
+                f"live instance {iid!r} is not on the approved fleet manifest "
+                "(unmanifested)"
+            ),
+            active_blocker=True,
+        )
+
+    # Same client_type multi-instance is healthy when each has distinct instance IDs
+    by_type: dict[str, list[str]] = defaultdict(list)
+    for inst in instances:
+        for ct in inst.get("client_types") or []:
+            by_type[str(ct)].append(inst["client_instance_id"])
+    multi_instance_same_type = {
+        ct: ids for ct, ids in by_type.items() if len(ids) > 1
+    }
+
+    active_blockers = [f for f in findings if f.get("active_blocker")]
+    historical_only = [f for f in findings if f.get("classification") == CLASS_HISTORICAL]
+    live_safe = not active_blockers
+
+    consistency = registry_revision or _consistency_token(all_rows, snapshot_at)
+
+    return {
+        "success": True,
+        "read_only": True,
+        "snapshot_at": snapshot_at,
+        "consistency_token": consistency,
+        "registry_revision": consistency,
+        "live_worker_count": len(live_rows),
+        "historical_worker_count": len(historical_rows),
+        "stale_worker_count": len(stale_rows),
+        "instance_count": len(instances),
+        "workers": all_rows,
+        "live_workers": live_rows,
+        "historical_workers": historical_rows,
+        "stale_workers": stale_rows,
+        "instances": instances,
+        "multi_instance_same_client_type": multi_instance_same_type,
+        "same_client_type_not_duplicate": True,
+        "expected_manifest": [
+            {
+                "client_instance_id": item.get("client_instance_id"),
+                "client_type": item.get("client_type"),
+                "fleet_run_id": item.get("fleet_run_id"),
+                "namespaces": item.get("namespaces"),
+            }
+            for item in expected
+        ],
+        "missing_expected_instance_ids": missing_ids,
+        "unmanifested_instance_ids": unmanifested_ids,
+        "findings": findings,
+        "active_blockers": active_blockers,
+        "historical_findings": historical_only,
+        "live_fleet_safe": live_safe,
+        "mutation_safe": live_safe and all(
+            inst.get("mutation_safe") for inst in instances
+        )
+        if instances
+        else live_safe,
+        "classification_model": {
+            "exactly_one_process_per_profile": False,
+            "exactly_one_instance_per_client_type": False,
+            "multiple_instances_per_client_type": True,
+            "duplicate_requires": [
+                "live client_instance_id collision",
+                "duplicate namespace worker within one instance",
+                "reused worker/session/generation/process/pid/fencing identity",
+                "cross-instance ownership collision",
+                "unmanifested instance when a manifest is required",
+            ],
+        },
+        "reasons": [f["detail"] for f in active_blockers],
+    }
+
+
+def compare_snapshot_heartbeats(
+    earlier: Mapping[str, Any],
+    later: Mapping[str, Any],
+) -> dict[str, Any]:
+    """Prove heartbeat continuity and stable ownership across two snapshots."""
+    earlier_live = {
+        w.get("worker_identity"): w for w in earlier.get("live_workers") or []
+    }
+    later_live = {
+        w.get("worker_identity"): w for w in later.get("live_workers") or []
+    }
+    shared = sorted(set(earlier_live) & set(later_live))
+    continuity: list[dict[str, Any]] = []
+    stable_ownership = True
+    for wid in shared:
+        a = earlier_live[wid]
+        b = later_live[wid]
+        same_instance = a.get("client_instance_id") == b.get("client_instance_id")
+        same_session = a.get("session_id") == b.get("session_id")
+        same_generation = a.get("generation_id") == b.get("generation_id")
+        hb_advanced_or_equal = True
+        if a.get("last_heartbeat_at") and b.get("last_heartbeat_at"):
+            hb_advanced_or_equal = b["last_heartbeat_at"] >= a["last_heartbeat_at"]
+        if not (same_instance and same_session and same_generation):
+            stable_ownership = False
+        continuity.append(
+            {
+                "worker_identity": wid,
+                "same_client_instance_id": same_instance,
+                "same_session_id": same_session,
+                "same_generation_id": same_generation,
+                "heartbeat_non_decreasing": hb_advanced_or_equal,
+                "earlier_heartbeat": a.get("last_heartbeat_at"),
+                "later_heartbeat": b.get("last_heartbeat_at"),
+            }
+        )
+    return {
+        "shared_live_workers": shared,
+        "continuity": continuity,
+        "stable_ownership": stable_ownership
+        and all(c["heartbeat_non_decreasing"] for c in continuity),
+        "dropped_workers": sorted(set(earlier_live) - set(later_live)),
+        "new_workers": sorted(set(later_live) - set(earlier_live)),
+    }

mcp_worker_identity.py

@@ -29,6 +29,7 @@ implements:
 
 from __future__ import annotations
 
+import atexit
 import hashlib
 import os
 import re
@@ -123,6 +124,103 @@ DEFAULT_REGISTRY_PATH = os.path.expanduser(
 #: worker within a single operator coffee break.
 DEFAULT_HEARTBEAT_TTL_SECONDS = 900.0
 
+#: Optional operator override for the beat interval, in seconds. Clamped by
+#: :func:`heartbeat_interval_for` so it can never be set at or above the TTL.
+HEARTBEAT_INTERVAL_ENV = "GITEA_WORKER_HEARTBEAT_INTERVAL_SECONDS"
+
+#: Never beat faster than this, so a misconfigured interval cannot turn the
+#: supervisor into a busy sqlite writer.
+MIN_HEARTBEAT_INTERVAL_SECONDS = 1.0
+
+
+def heartbeat_interval_for(
+    ttl_seconds: float = DEFAULT_HEARTBEAT_TTL_SECONDS,
+    override: str | float | None = None,
+) -> float:
+    """Beat interval for *ttl_seconds*: one third of the TTL, capped at a half.
+
+    #975. A third means two consecutive beats can be lost before the row is
+    allowed to look stale, and the half-TTL cap is a hard ceiling so no
+    override can produce an interval that expires the registration it is
+    supposed to renew. That is the whole liveness contract: a healthy worker
+    stays owned, and a worker that has genuinely stopped beating still expires
+    on the configured TTL — this function never touches the TTL itself.
+    """
+    try:
+        ttl = float(ttl_seconds)
+    except (TypeError, ValueError):
+        ttl = DEFAULT_HEARTBEAT_TTL_SECONDS
+    if not ttl > 0:
+        ttl = DEFAULT_HEARTBEAT_TTL_SECONDS
+
+    interval = ttl / 3.0
+    if override is not None:
+        candidate = str(override).strip()
+        if candidate:
+            try:
+                parsed = float(candidate)
+            except (TypeError, ValueError):
+                parsed = None
+            if parsed is not None and parsed > 0:
+                interval = parsed
+
+    ceiling = ttl / 2.0
+    if interval > ceiling:
+        interval = ceiling
+    if interval < MIN_HEARTBEAT_INTERVAL_SECONDS:
+        interval = min(MIN_HEARTBEAT_INTERVAL_SECONDS, ceiling)
+    return interval
+
+
+#: Recorded-vs-presented pairs compared by :func:`_heartbeat_expectation_drift`.
+_HEARTBEAT_EXPECTATION_FIELDS = (
+    "session_id",
+    "generation_id",
+    "client_name",
+    "pid",
+)
+
+
+def _heartbeat_expectation_drift(
+    record: dict[str, Any],
+    *,
+    expected_session_id: str | None = None,
+    expected_generation_id: str | None = None,
+    expected_client_name: str | None = None,
+    expected_pid: int | None = None,
+) -> list[tuple[str, Any, Any]]:
+    """Return ``(field, recorded, presented)`` for every mismatched expectation.
+
+    Only supplied expectations are compared, so a caller that presents nothing
+    gets the pre-#975 behaviour. ``client_name`` is compared through
+    :func:`normalize_client_name` so one application's namespaces stay one
+    client while genuinely different clients stay distinct.
+    """
+    presented = {
+        "session_id": expected_session_id,
+        "generation_id": expected_generation_id,
+        "client_name": (
+            normalize_client_name(expected_client_name)
+            if expected_client_name is not None
+            else None
+        ),
+        "pid": expected_pid,
+    }
+    drift: list[tuple[str, Any, Any]] = []
+    for field in _HEARTBEAT_EXPECTATION_FIELDS:
+        want = presented[field]
+        if want is None:
+            continue
+        got = record.get(field)
+        if field == "pid":
+            match = got is not None and int(got) == int(want)
+        else:
+            match = got == want
+        if not match:
+            drift.append((field, got, want))
+    return drift
+
+
 STATUS_ACTIVE = "active"
 STATUS_SUPERSEDED = "superseded"
 STATUS_RELEASED = "released"
@@ -203,8 +301,22 @@ CREATE INDEX IF NOT EXISTS idx_worker_session
     ON worker_registrations(session_id, status);
 CREATE INDEX IF NOT EXISTS idx_worker_profile
     ON worker_registrations(profile, status);
+CREATE INDEX IF NOT EXISTS idx_worker_instance
+    ON worker_registrations(client_instance_id, status);
 """
 
+#: #978 optional columns added without rewriting historical rows.
+_SCHEMA_OPTIONAL_COLUMNS: tuple[tuple[str, str], ...] = (
+    ("fleet_run_id", "TEXT"),
+    ("authenticated_account", "TEXT"),
+    ("process_identity", "TEXT"),
+    ("startup_revision", "TEXT"),
+    ("loaded_revision", "TEXT"),
+    ("parity_revision", "TEXT"),
+    ("live_revision", "TEXT"),
+    ("instance_id_provenance", "TEXT"),
+)
+
 
 class WorkerRegistryError(RuntimeError):
     """Raised for registry misuse that is a programming error, not a refusal."""
@@ -536,6 +648,17 @@ class WorkerRegistry:
             conn = self._connect()
             try:
                 conn.executescript(_SCHEMA_SQL)
+                existing = {
+                    row[1]
+                    for row in conn.execute(
+                        "PRAGMA table_info(worker_registrations)"
+                    ).fetchall()
+                }
+                for name, decl in _SCHEMA_OPTIONAL_COLUMNS:
+                    if name not in existing:
+                        conn.execute(
+                            f"ALTER TABLE worker_registrations ADD COLUMN {name} {decl}"
+                        )
                 conn.commit()
             finally:
                 conn.close()
@@ -660,6 +783,14 @@ class WorkerRegistry:
         heartbeat_ttl_seconds: float = DEFAULT_HEARTBEAT_TTL_SECONDS,
         now: datetime | None = None,
         pid_alive_probe=None,
+        fleet_run_id: str | None = None,
+        authenticated_account: str | None = None,
+        process_identity: str | None = None,
+        startup_revision: str | None = None,
+        loaded_revision: str | None = None,
+        parity_revision: str | None = None,
+        live_revision: str | None = None,
+        instance_id_provenance: str | None = None,
     ) -> dict[str, Any]:
         """Atomically register one worker identity.
 
@@ -668,6 +799,10 @@ class WorkerRegistry:
         to mint a different identity and register that instead (AC32). The
         existing registration is returned untouched so the caller can see what
         it collided with.
+
+        #978 also fails closed when a *live* registration already holds the same
+        ``client_instance_id`` for the same namespace: two workers of one
+        namespace cannot share one instance.
         """
         parsed = parse_worker_identity(worker_identity)
         if not parsed["valid"]:
@@ -685,6 +820,9 @@ class WorkerRegistry:
             }
 
         stamp = _ts(now or _utc_now())
+        proc_id = process_identity or (
+            f"pid-{int(pid)}" if pid is not None else None
+        )
         with self._tx() as conn:
             existing = conn.execute(
                 "SELECT * FROM worker_registrations WHERE worker_identity = ?",
@@ -721,6 +859,41 @@ class WorkerRegistry:
                     ),
                 }
 
+            # #978: refuse a second live worker for the same (instance, namespace).
+            if namespace and client_instance_id:
+                peers = conn.execute(
+                    "SELECT * FROM worker_registrations "
+                    "WHERE client_instance_id = ? AND namespace = ? AND status = ?",
+                    (client_instance_id, namespace, STATUS_ACTIVE),
+                ).fetchall()
+                for peer_row in peers:
+                    peer = self._row_to_record(peer_row)
+                    pid_alive = (
+                        pid_alive_probe(peer.get("pid"))
+                        if pid_alive_probe is not None
+                        else None
+                    )
+                    if self.is_live(peer, now=now, pid_alive=pid_alive)["live"]:
+                        return {
+                            "success": False,
+                            "registered": False,
+                            "mutation_performed": False,
+                            "blocker_kind": BLOCKER_CONFLICTING_SESSIONS,
+                            "collision": True,
+                            "collision_kind": "duplicate_namespace_worker",
+                            "existing_registration": _public_record(peer),
+                            "reasons": [
+                                f"client_instance_id {client_instance_id!r} already "
+                                f"has a live worker for namespace {namespace!r} "
+                                f"({peer.get('worker_identity')!r}); #978 fails closed "
+                                "rather than registering a second worker"
+                            ],
+                            "exact_next_action": (
+                                "Stop the extra namespace worker, or use a distinct "
+                                "client_instance_id for a separate application launch."
+                            ),
+                        }
+
             epoch = self._next_epoch(conn, generation_id)
             conn.execute(
                 """
@@ -729,8 +902,11 @@ class WorkerRegistry:
                     generation_id, role, profile, namespace, remote,
                     repository_binding, pid, transport, token_fingerprint,
                     started_at, last_heartbeat_at, heartbeat_ttl_seconds,
-                    fencing_epoch, status
-                ) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)
+                    fencing_epoch, status,
+                    fleet_run_id, authenticated_account, process_identity,
+                    startup_revision, loaded_revision, parity_revision,
+                    live_revision, instance_id_provenance
+                ) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)
                 """,
                 (
                     worker_identity,
@@ -751,6 +927,14 @@ class WorkerRegistry:
                     float(heartbeat_ttl_seconds),
                     epoch,
                     STATUS_ACTIVE,
+                    (fleet_run_id or "").strip() or None,
+                    (authenticated_account or "").strip() or None,
+                    proc_id,
+                    (startup_revision or "").strip() or None,
+                    (loaded_revision or "").strip() or None,
+                    (parity_revision or "").strip() or None,
+                    (live_revision or "").strip() or None,
+                    (instance_id_provenance or "").strip() or None,
                 ),
             )
             row = conn.execute(
@@ -786,11 +970,25 @@ class WorkerRegistry:
         worker_identity: str,
         fencing_epoch: int,
         now: datetime | None = None,
+        expected_session_id: str | None = None,
+        expected_generation_id: str | None = None,
+        expected_client_name: str | None = None,
+        expected_pid: int | None = None,
     ) -> dict[str, Any]:
         """Renew only the owning registration (#948 AC11).
 
         A stale epoch is refused rather than silently renewed, so a superseded
         session that resumes cannot heartbeat its way back into ownership.
+
+        #975 adds optional keyword-only *expectations*. Every one supplied must
+        match the recorded row or the renewal is refused. They exist because
+        identity plus epoch cannot express "renew the row I registered, and only
+        that row": a recycled PID, or a second session of the same client, would
+        otherwise be renewable by the wrong beater. They are fencing tokens,
+        never assertions — supplying one can only cause a refusal, never grant
+        anything, and omitting them preserves the pre-#975 behaviour exactly.
+        Drift reports the existing ``BLOCKER_FENCED`` rather than a new
+        ``blocker_kind``, because consumers switch on that value.
         """
         stamp = _ts(now or _utc_now())
         with self._tx() as conn:
@@ -807,6 +1005,30 @@ class WorkerRegistry:
                     "reasons": [f"no registration for {worker_identity!r}"],
                 }
             record = self._row_to_record(row)
+            drift = _heartbeat_expectation_drift(
+                record,
+                expected_session_id=expected_session_id,
+                expected_generation_id=expected_generation_id,
+                expected_client_name=expected_client_name,
+                expected_pid=expected_pid,
+            )
+            if drift:
+                return {
+                    "success": False,
+                    "renewed": False,
+                    "mutation_performed": False,
+                    "blocker_kind": BLOCKER_FENCED,
+                    "expectation_drift": drift,
+                    "reasons": [
+                        "presented worker expectations do not match the recorded "
+                        "registration, so this beater does not own the row: "
+                        + "; ".join(
+                            f"{field} recorded {recorded!r}, presented {presented!r}"
+                            for field, recorded, presented in drift
+                        )
+                        + " (#975)"
+                    ],
+                }
             if record["status"] != STATUS_ACTIVE:
                 return {
                     "success": False,
@@ -1013,8 +1235,17 @@ def _public_record(record: dict[str, Any]) -> dict[str, Any]:
         "transport": record.get("transport"),
         "started_at": record.get("started_at"),
         "last_heartbeat_at": record.get("last_heartbeat_at"),
+        "heartbeat_ttl_seconds": record.get("heartbeat_ttl_seconds"),
         "fencing_epoch": record.get("fencing_epoch"),
         "status": record.get("status"),
+        "fleet_run_id": record.get("fleet_run_id"),
+        "authenticated_account": record.get("authenticated_account"),
+        "process_identity": record.get("process_identity"),
+        "startup_revision": record.get("startup_revision"),
+        "loaded_revision": record.get("loaded_revision"),
+        "parity_revision": record.get("parity_revision"),
+        "live_revision": record.get("live_revision"),
+        "instance_id_provenance": record.get("instance_id_provenance"),
     }
 
 
@@ -1430,3 +1661,253 @@ def resolve_bound_remote(
             "explicitly to avoid host drift (#948)."
         ],
     }
+
+
+# --- Production heartbeat lifecycle (#975) --------------------------------
+#
+# #948 delivered ``WorkerRegistry.heartbeat()`` and nothing ever called it, so
+# ``last_heartbeat_at`` stayed pinned to ``started_at`` for every production
+# worker and ``heartbeat_ttl_seconds`` stopped being a liveness window at all —
+# it became a hard cap on how long any client could stay attached. This
+# supervisor is the missing caller.
+#
+# It is a daemon thread rather than an asyncio task or a per-request refresh
+# because the renewal has to survive an *idle* session: a client waiting on a
+# lease with no tool call in flight must not lose ownership, which rules out
+# request-driven refresh. Registration itself happens lazily inside a tool call
+# on the server's event loop, and the registry performs blocking
+# ``BEGIN IMMEDIATE`` sqlite writes, which must not run on that loop.
+# ``daemon=True`` is deliberate: a hard kill takes the thread down with the
+# process, so a dead worker still goes stale on the normal TTL.
+
+#: Refusals that mean this worker no longer owns its row. Beating again could
+#: only ever be an attempt to renew ownership it has already lost, so the
+#: supervisor stops permanently instead of retrying.
+TERMINAL_HEARTBEAT_BLOCKERS = frozenset({BLOCKER_FENCED, BLOCKER_NO_ATTACHMENT})
+
+
+class WorkerHeartbeatSupervisor:
+    """Periodically renew exactly one worker registration.
+
+    Contract:
+
+    * It never registers. A supervisor exists only for an already-registered
+      identity, so it cannot create a second registration or a second identity
+      system.
+    * Every beat presents the full expectation set, so it can renew only the row
+      matching this exact client name, worker identity, session, generation and
+      pid.
+    * A terminal refusal (fenced or missing) stops it permanently and records
+      why. A fenced session must never beat its way back into ownership.
+    * A transient failure (a locked database, say) is counted and the loop
+      continues, so one contended write does not silently end the heartbeat.
+    * Nothing here raises into a caller. ``beat_once`` returns its outcome and
+      the thread body swallows everything, because a heartbeat failure must
+      degrade to "not renewed" and never crash a tool call.
+    """
+
+    def __init__(
+        self,
+        registry: WorkerRegistry,
+        *,
+        worker_identity: str,
+        fencing_epoch: int,
+        session_id: str | None = None,
+        generation_id: str | None = None,
+        client_name: str | None = None,
+        pid: int | None = None,
+        ttl_seconds: float = DEFAULT_HEARTBEAT_TTL_SECONDS,
+        interval_seconds: float | None = None,
+        clock=None,
+    ) -> None:
+        self._registry = registry
+        self.worker_identity = worker_identity
+        self.fencing_epoch = int(fencing_epoch)
+        self.session_id = session_id
+        self.generation_id = generation_id
+        self.client_name = (
+            normalize_client_name(client_name) if client_name is not None else None
+        )
+        self.pid = int(pid) if pid is not None else None
+        self.ttl_seconds = float(ttl_seconds)
+        self.interval_seconds = (
+            heartbeat_interval_for(self.ttl_seconds)
+            if interval_seconds is None
+            else heartbeat_interval_for(self.ttl_seconds, interval_seconds)
+        )
+        #: Injectable so every TTL test uses controlled time and no test waits
+        #: for a real interval or a real TTL to elapse.
+        self._clock = clock or _utc_now
+
+        self._stop_event = threading.Event()
+        self._thread: threading.Thread | None = None
+        self._lock = threading.Lock()
+        self._atexit_registered = False
+
+        self.started = False
+        self.stopped_reason: str | None = None
+        self.beats_attempted = 0
+        self.beats_renewed = 0
+        self.transient_failures = 0
+        self.last_beat_at: str | None = None
+        self.last_result: dict[str, Any] | None = None
+
+    # -- one beat --
+
+    def beat_once(self, now: datetime | None = None) -> dict[str, Any]:
+        """Renew once. Never raises; returns the registry outcome or a failure."""
+        if self.stopped_reason is not None:
+            return {
+                "success": False,
+                "renewed": False,
+                "beat_attempted": False,
+                "reasons": [
+                    f"supervisor already stopped: {self.stopped_reason}"
+                ],
+            }
+
+        stamp = now or self._clock()
+        self.beats_attempted += 1
+        try:
+            result = self._registry.heartbeat(
+                worker_identity=self.worker_identity,
+                fencing_epoch=self.fencing_epoch,
+                now=stamp,
+                expected_session_id=self.session_id,
+                expected_generation_id=self.generation_id,
+                expected_client_name=self.client_name,
+                expected_pid=self.pid,
+            )
+        except Exception as exc:
+            # Transient by assumption: an unexpected error is not proof this
+            # worker lost ownership, so it must not silently end the heartbeat.
+            self.transient_failures += 1
+            result = {
+                "success": False,
+                "renewed": False,
+                "transient": True,
+                "blocker_kind": None,
+                "reasons": [f"heartbeat raised {type(exc).__name__}: {exc}"],
+            }
+            self.last_result = result
+            return result
+
+        self.last_result = result
+        if result.get("renewed"):
+            self.beats_renewed += 1
+            self.last_beat_at = result.get("last_heartbeat_at") or _ts(stamp)
+            return result
+
+        blocker = result.get("blocker_kind")
+        if blocker in TERMINAL_HEARTBEAT_BLOCKERS:
+            self._stop_internal(
+                reason=(
+                    f"refused with blocker_kind={blocker!r}: "
+                    + "; ".join(result.get("reasons") or [])
+                )
+            )
+        else:
+            self.transient_failures += 1
+        return result
+
+    # -- lifecycle --
+
+    def start(self) -> dict[str, Any]:
+        """Start the beat thread. Idempotent; safe to call from a tool call."""
+        with self._lock:
+            if self.stopped_reason is not None:
+                return {
+                    "started": False,
+                    "reasons": [f"supervisor stopped: {self.stopped_reason}"],
+                }
+            if self._thread is not None and self._thread.is_alive():
+                return {"started": True, "already_running": True, "reasons": []}
+
+            self._stop_event.clear()
+            thread = threading.Thread(
+                target=self._run,
+                name=f"gitea-worker-heartbeat-{self.worker_identity}",
+                daemon=True,
+            )
+            self._thread = thread
+            self.started = True
+            if not self._atexit_registered:
+                # Orderly shutdown stops the heartbeat. A hard kill does not
+                # run this, which is correct: the row must then go stale.
+                atexit.register(self._atexit_stop)
+                self._atexit_registered = True
+            thread.start()
+            return {"started": True, "already_running": False, "reasons": []}
+
+    def _run(self) -> None:
+        while not self._stop_event.is_set():
+            # Wait first: registration already stamped a fresh heartbeat, so an
+            # immediate beat would be a redundant write on every server launch.
+            if self._stop_event.wait(self.interval_seconds):
+                return
+            try:
+                self.beat_once()
+            except Exception:
+                # beat_once is already total; this is the last-resort guard that
+                # keeps a supervisor thread from dying silently.
+                self.transient_failures += 1
+            if self.stopped_reason is not None:
+                return
+
+    def stop(self, reason: str = "stopped") -> dict[str, Any]:
+        """Stop beating. Idempotent, and prompt because the loop waits on an Event."""
+        self._stop_internal(reason=reason)
+        thread = self._thread
+        if thread is not None and thread.is_alive():
+            thread.join(timeout=max(1.0, min(5.0, self.interval_seconds)))
+        return {
+            "stopped": True,
+            "reason": self.stopped_reason,
+            "thread_alive": bool(thread is not None and thread.is_alive()),
+        }
+
+    def _stop_internal(self, *, reason: str) -> None:
+        if self.stopped_reason is None:
+            self.stopped_reason = reason
+        self._stop_event.set()
+        if self._atexit_registered:
+            try:
+                atexit.unregister(self._atexit_stop)
+            except Exception:
+                pass
+            self._atexit_registered = False
+
+    def _atexit_stop(self) -> None:
+        try:
+            self.stop(reason="process exit")
+        except Exception:
+            pass
+
+    # -- observability --
+
+    def status(self) -> dict[str, Any]:
+        """Read-only observability payload; safe to embed in a tool result."""
+        thread = self._thread
+        return {
+            "supervised": True,
+            "worker_identity": self.worker_identity,
+            "fencing_epoch": self.fencing_epoch,
+            "session_id": self.session_id,
+            "generation_id": self.generation_id,
+            "client_name": self.client_name,
+            "pid": self.pid,
+            "heartbeat_ttl_seconds": self.ttl_seconds,
+            "heartbeat_interval_seconds": self.interval_seconds,
+            "started": self.started,
+            "running": bool(
+                thread is not None
+                and thread.is_alive()
+                and self.stopped_reason is None
+            ),
+            "stopped_reason": self.stopped_reason,
+            "beats_attempted": self.beats_attempted,
+            "beats_renewed": self.beats_renewed,
+            "transient_failures": self.transient_failures,
+            "last_heartbeat_at": self.last_beat_at,
+            "last_blocker_kind": (self.last_result or {}).get("blocker_kind"),
+        }

root_checkout_guard.py

@@ -1,9 +1,16 @@
 """Root checkout guard (#475).
 
-The project root checkout is the stable control checkout on master/prgs/master.
-Author/reviewer/merge flows must fail closed when the control checkout is
-contaminated (wrong branch, detached HEAD, dirty, or HEAD behind/ahead of
-prgs/master). Isolated ``branches/...`` worktrees remain allowed.
+The project root checkout is the stable control checkout on its integration
+branch. Author/reviewer/merge flows must fail closed when the control checkout
+is contaminated (wrong branch, detached HEAD, dirty, or HEAD behind/ahead of the
+tracking integration ref). Isolated ``branches/...`` worktrees remain allowed.
+
+The tracking ref is *derived per repository* (#983) rather than assumed to be
+``prgs/master``: a namespace bound to another repository — say remote ``MDCPS``
+on branch ``dev`` — is gated against ``refs/remotes/MDCPS/dev``. Derivation is
+delegated to :mod:`canonical_repository_root`, the authoritative
+repository/context resolver, so gating and parity reporting share one resolved
+target instead of maintaining two disagreeing hardcoded defaults.
 """
 
 from __future__ import annotations
@@ -11,6 +18,7 @@ from __future__ import annotations
 import os
 import subprocess
 
+import canonical_repository_root
 from author_mutation_worktree import is_path_under_branches
 from reviewer_worktree import parse_dirty_tracked_files
 
@@ -20,19 +28,65 @@ REMEDIATION = (
 )
 
 BASE_BRANCHES = frozenset({"master", "main", "dev"})
+
+# Legacy PRGS-specific probe order. Retained only for callers that pass an
+# explicit ``remote_refs`` override; it is no longer the silent default, because
+# inheriting it in a non-PRGS checkout compared that checkout against a ref it
+# can never have (#983).
 REMOTE_MASTER_REFS = ("prgs/master", "refs/remotes/prgs/master")
 
 
+def _derive_probe_refs(root: str, explicit_remote: str | None) -> dict:
+    """Derive the ordered tracking refs to probe for *root*.
+
+    Returns the derivation payload from :mod:`canonical_repository_root` plus a
+    ``refs`` tuple, which is empty when the target is not provable.
+    """
+    derived = canonical_repository_root.resolve_target_base_ref(
+        root, explicit_remote=explicit_remote
+    )
+    return {
+        "refs": tuple(derived.get("tracking_refs") or ()),
+        "remote": derived.get("remote"),
+        "branch": derived.get("branch"),
+        "source": derived.get("source"),
+        "proven": bool(derived.get("proven")),
+        "reason_code": derived.get("reason_code"),
+        "reasons": list(derived.get("reasons") or []),
+        "cached_remote_head_branch": derived.get("cached_remote_head_branch"),
+        "cached_remote_head_conflicts": bool(derived.get("cached_remote_head_conflicts")),
+    }
+
+
 def resolve_remote_master_sha(
     canonical_repo_root: str,
     *,
     remote_refs: tuple[str, ...] | None = None,
+    explicit_remote: str | None = None,
 ) -> str | None:
-    """Return the commit SHA for the tracking master ref when available."""
+    """Return the commit SHA for the tracking integration ref when available.
+
+    This remains the single place that turns a ref into a SHA. Returns None when
+    the target cannot be derived or resolved — exactly what this function already
+    returned when ``rev-parse`` failed. Callers that must fail closed on missing
+    evidence (the #749/#757 bootstrap path) surface that None as *missing
+    evidence*, never as "no constraint".
+
+    *explicit_remote* is a caller-supplied disambiguation and is named that way
+    deliberately: an internally inferred remote handed back in would suppress the
+    resolver's ambiguity gate (#983 B2). No production caller supplies it.
+    """
     root = (canonical_repo_root or "").strip()
     if not root:
         return None
-    for ref in remote_refs or REMOTE_MASTER_REFS:
+    if remote_refs:
+        probe: tuple[str, ...] = tuple(remote_refs)
+    else:
+        derived = _derive_probe_refs(root, explicit_remote)
+        if not derived["proven"]:
+            return None
+        probe = derived["refs"]
+    for ref in probe:
         res = subprocess.run(
             ["git", "-C", root, "rev-parse", "--verify", ref],
             capture_output=True,
@@ -46,6 +100,84 @@ def resolve_remote_master_sha(
     return None
 
 
+def resolve_remote_master_ref_state(
+    canonical_repo_root: str,
+    *,
+    remote_refs: tuple[str, ...] | None = None,
+    explicit_remote: str | None = None,
+) -> dict:
+    """Resolve the tracking integration ref together with its commit SHA.
+
+    Returns ``sha``, the ``ref`` it came from, the derived ``remote`` /
+    ``branch``, a machine-checkable ``reason_code``, and ``reasons``. ``sha`` is
+    None whenever the target cannot be resolved — never a fallback to some other
+    repository's commit.
+
+    The SHA itself is obtained through :func:`resolve_remote_master_sha` rather
+    than by re-probing here, so one public function stays authoritative for
+    ref-to-SHA resolution. An explicit *remote_refs* keeps the historical
+    behaviour exactly: those refs are probed in order and no derivation happens.
+    """
+    root = (canonical_repo_root or "").strip()
+    state: dict = {
+        "sha": None,
+        "ref": None,
+        "remote": None,
+        "branch": None,
+        "source": None,
+        "reason_code": None,
+        "reasons": [],
+        "cached_remote_head_branch": None,
+        "cached_remote_head_conflicts": False,
+    }
+    if not root:
+        state["reasons"].append("no canonical repository root supplied (fail closed)")
+        return state
+
+    if remote_refs:
+        probe: tuple[str, ...] = tuple(remote_refs)
+        state["source"] = "explicit_remote_refs"
+    else:
+        derived = _derive_probe_refs(root, explicit_remote)
+        state["remote"] = derived["remote"]
+        state["branch"] = derived["branch"]
+        state["source"] = derived["source"]
+        state["cached_remote_head_branch"] = derived["cached_remote_head_branch"]
+        state["cached_remote_head_conflicts"] = derived["cached_remote_head_conflicts"]
+        if not derived["proven"]:
+            state["reason_code"] = derived["reason_code"]
+            state["reasons"] = derived["reasons"]
+            return state
+        probe = derived["refs"]
+
+    sha = resolve_remote_master_sha(root, remote_refs=probe, explicit_remote=explicit_remote)
+    if not sha:
+        state["reasons"].append(
+            "tracking integration ref "
+            f"{' / '.join(probe) if probe else '(none derived)'} does not resolve in "
+            f"'{root}' (fail closed)"
+        )
+        return state
+
+    state["sha"] = sha
+    # Name the ref that actually carries this commit. When the SHA comes from a
+    # test double no probe will match, so fall back to the first derived ref,
+    # which is the one the guard is conceptually comparing against.
+    for ref in probe:
+        res = subprocess.run(
+            ["git", "-C", root, "rev-parse", "--verify", ref],
+            capture_output=True,
+            text=True,
+            check=False,
+        )
+        if res.returncode == 0 and (res.stdout or "").strip() == sha:
+            state["ref"] = ref
+            break
+    else:
+        state["ref"] = probe[0] if probe else None
+    return state
+
+
 resolve_tracking_master_sha = resolve_remote_master_sha
 
 
@@ -59,8 +191,9 @@ def assess_root_checkout_guard(
     remote_master_sha: str | None,
     resolved_role: str | None = None,
     actual_role: str | None = None,
+    remote_master_ref: str | None = None,
 ) -> dict:
-    """Fail closed when the control checkout is not clean master/prgs/master.
+    """Fail closed when the control checkout is not clean on its integration ref.
 
     ``resolved_role`` is the preflight-resolved *task* role and ``actual_role``
     is the *active profile* role (#540). The reconciler exemption honours either
@@ -102,9 +235,13 @@ def assess_root_checkout_guard(
         )
 
     if remote_master_sha and head_sha and head_sha != remote_master_sha:
+        # #983: name the ref that was actually compared. Reporting a literal
+        # 'prgs/master' in a checkout gated against refs/remotes/MDCPS/dev sends
+        # the operator to inspect a ref that repository does not have.
+        ref_label = (remote_master_ref or "").strip() or "the tracking integration ref"
         reasons.append(
-            "control checkout HEAD does not match prgs/master "
-            f"(HEAD {head_sha[:12]}, prgs/master {remote_master_sha[:12]})"
+            f"control checkout HEAD does not match {ref_label} "
+            f"(HEAD {head_sha[:12]}, {ref_label} {remote_master_sha[:12]})"
         )
 
     proven = not reasons

task_capability_map.py

@@ -163,6 +163,18 @@ TASK_CAPABILITY_MAP: dict[str, dict[str, str]] = {
         "permission": "gitea.read",
         "role": "author",
     },
+    # #978: instance-level fleet identity/health snapshot. gitea.read is the
+    # operation gate; the tool additionally restricts role_kind to
+    # controller|reconciler so author/reviewer/merger cannot use it as a
+    # mutation surface and no unrelated write permission is introduced.
+    "snapshot_instance_fleet": {
+        "permission": "gitea.read",
+        "role": "controller",
+    },
+    "gitea_snapshot_instance_fleet": {
+        "permission": "gitea.read",
+        "role": "controller",
+    },
     # #644: Phase 2 Web Console recovery tasks.
     "clear_stale_binding": {
         "permission": "gitea.read",

tests/test_config_menu.py

@@ -175,8 +175,21 @@ class TestLauncherSnippets(unittest.TestCase):
     def test_only_safe_keys_no_secrets(self):
         entry = gitea_config.launcher_entry("prgs", "/cfg/profiles.json")["gitea-tools"]
         self.assertEqual(set(entry), {"command", "args", "env"})
-        self.assertEqual(set(entry["env"]), {"GITEA_MCP_CONFIG", "GITEA_MCP_PROFILE", "GITEA_CLIENT_MANAGED"})
+        # #978 B1: production launcher also injects trusted client identity.
+        required = {
+            "GITEA_MCP_CONFIG",
+            "GITEA_MCP_PROFILE",
+            "GITEA_CLIENT_MANAGED",
+            "GITEA_MCP_CLIENT",
+            "GITEA_MCP_CLIENT_INSTANCE",
+            "GITEA_MCP_INSTANCE_PROVENANCE",
+        }
+        self.assertTrue(required.issubset(set(entry["env"])), entry["env"])
         self.assertEqual(entry["env"]["GITEA_MCP_PROFILE"], "prgs")
+        self.assertTrue(
+            entry["env"]["GITEA_MCP_CLIENT_INSTANCE"].startswith("inst-"),
+            entry["env"]["GITEA_MCP_CLIENT_INSTANCE"],
+        )
         blob = json.dumps(entry).lower()
         for word in ("token", "password", "secret"):
             self.assertNotIn(word, blob)

tests/test_issue_983_cross_repo_base_ref.py

@@ -0,0 +1,692 @@
+"""Regression tests for Issue #983: derived target base ref for cross-repository checkouts.
+
+The mutation guard previously assumed ``prgs/master`` and the parity report
+assumed ``origin/master``. Any repository using neither — for example remote
+``MDCPS`` on integration branch ``dev`` — could not prove base equivalence, so
+every gated mutation failed closed with no reachable remedy.
+
+Two further defects were found by review at head ``2d5d5c9d`` and are covered
+here:
+
+* **B1** — the first fix derived the integration branch from
+  ``refs/remotes/<remote>/HEAD``. That symref is a *local cache* written at clone
+  time and never refreshed by fetch, so on the real Weekly Briefings target it
+  still named ``main`` while the checkout tracked and sat exactly on ``dev``. The
+  authoritative signal is the checkout's own configured upstream. The fixtures
+  below therefore reproduce the **disagreement**: the cache says ``main``, the
+  configured upstream says ``dev``, and ``dev`` must win.
+* **B2** — the parity report passed its internally inferred identity remote back
+  into the resolver, which reads a caller-supplied remote as "the caller already
+  disambiguated" and skips its ambiguity gate. Gating and reporting could then
+  evaluate different repositories. An inferred remote is now never laundered into
+  explicit caller intent, and ambiguity fails closed on both sides.
+
+These tests build hermetic git repositories on disk (no network, no fetch) and
+assert the derived target end to end: identity remote, integration branch,
+tracking ref, fail-closed refusals, and agreement between the mutation guard and
+the parity report.
+"""
+
+from __future__ import annotations
+
+import inspect
+import os
+import subprocess
+import tempfile
+import unittest
+
+import anti_stomp_preflight
+import canonical_repository_root as crr
+import master_parity_gate
+import root_checkout_guard
+
+MDCPS_URL = "https://gitea.example.net/MDCPS/WeeklyBriefings-Meta.git"
+MDCPS_SLUG = "MDCPS/WeeklyBriefings-Meta"
+PRGS_URL = "https://gitea.prgs.cc/Scaled-Tech-Consulting/Gitea-Tools.git"
+PRGS_SLUG = "Scaled-Tech-Consulting/Gitea-Tools"
+
+
+def _git(root: str, *args: str) -> str:
+    res = subprocess.run(
+        ["git", "-C", root, *args],
+        capture_output=True,
+        text=True,
+        check=True,
+    )
+    return (res.stdout or "").strip()
+
+
+def _make_repo(root: str, *, remote: str | None, url: str | None) -> str:
+    """Initialise a repository with one commit and an optional named remote."""
+    os.makedirs(root, exist_ok=True)
+    _git(root, "init", "--quiet")
+    _git(root, "config", "user.email", "[email protected]")
+    _git(root, "config", "user.name", "Issue983 Test")
+    _git(root, "config", "commit.gpgsign", "false")
+    with open(os.path.join(root, "seed.txt"), "w", encoding="utf-8") as fh:
+        fh.write("seed\n")
+    _git(root, "add", "seed.txt")
+    _git(root, "commit", "--quiet", "-m", "seed")
+    if remote and url:
+        _git(root, "remote", "add", remote, url)
+    return _git(root, "rev-parse", "HEAD")
+
+
+def _set_remote_branch(root: str, remote: str, branch: str, sha: str) -> None:
+    """Create refs/remotes/<remote>/<branch> without contacting a network."""
+    _git(root, "update-ref", f"refs/remotes/{remote}/{branch}", sha)
+
+
+def _set_remote_head(root: str, remote: str, branch: str) -> None:
+    """Write the *cached* refs/remotes/<remote>/HEAD symref.
+
+    This is the signal B1 proved untrustworthy. Fixtures use it to reproduce a
+    stale cache, never to manufacture the answer under test.
+    """
+    _git(
+        root,
+        "symbolic-ref",
+        f"refs/remotes/{remote}/HEAD",
+        f"refs/remotes/{remote}/{branch}",
+    )
+
+
+def _set_upstream(root: str, remote: str, branch: str) -> None:
+    """Configure the current branch's upstream, exactly as git tracking does.
+
+    Writes ``branch.<current>.remote`` / ``branch.<current>.merge`` directly
+    rather than via ``--set-upstream-to`` so no ref is required to pre-exist and
+    no network is touched.
+    """
+    current = _git(root, "symbolic-ref", "--short", "HEAD")
+    _git(root, "config", f"branch.{current}.remote", remote)
+    _git(root, "config", f"branch.{current}.merge", f"refs/heads/{branch}")
+
+
+def _checkout_new_branch(root: str, branch: str) -> None:
+    _git(root, "checkout", "--quiet", "-b", branch)
+
+
+def _advance(root: str, message: str) -> str:
+    with open(os.path.join(root, "seed.txt"), "a", encoding="utf-8") as fh:
+        fh.write(message + "\n")
+    _git(root, "add", "seed.txt")
+    _git(root, "commit", "--quiet", "-m", message)
+    return _git(root, "rev-parse", "HEAD")
+
+
+class _RepoCase(unittest.TestCase):
+    def setUp(self) -> None:
+        self._tmp = tempfile.TemporaryDirectory()
+        self.addCleanup(self._tmp.cleanup)
+        self.root = os.path.join(self._tmp.name, "repo")
+
+    def _weekly_briefings_shape(self) -> str:
+        """The real Weekly Briefings target, including its stale cache.
+
+        Remote ``MDCPS``; checked out on ``dev``; upstream configured to
+        ``MDCPS/dev``; both ``dev`` and ``main`` present as tracking refs; and
+        ``refs/remotes/MDCPS/HEAD`` still cached at ``main`` from clone time.
+        """
+        head = _make_repo(self.root, remote="MDCPS", url=MDCPS_URL)
+        _checkout_new_branch(self.root, "dev")
+        _set_remote_branch(self.root, "MDCPS", "dev", head)
+        stale = _advance(self.root, "main diverged long ago")
+        _set_remote_branch(self.root, "MDCPS", "main", stale)
+        _git(self.root, "reset", "--hard", "--quiet", head)
+        _set_remote_head(self.root, "MDCPS", "main")  # stale clone-time cache
+        _set_upstream(self.root, "MDCPS", "dev")  # authoritative
+        return head
+
+
+class TestPrgsBehaviourPreserved(_RepoCase):
+    """Required coverage 1: PRGS prgs/master compatibility."""
+
+    def _prgs(self) -> str:
+        head = _make_repo(self.root, remote="prgs", url=PRGS_URL)
+        _set_remote_branch(self.root, "prgs", "master", head)
+        _set_remote_head(self.root, "prgs", "master")
+        _set_upstream(self.root, "prgs", "master")
+        return head
+
+    def test_prgs_master_resolves_unchanged(self):
+        head = self._prgs()
+
+        got = crr.resolve_target_base_ref(self.root)
+        self.assertTrue(got["proven"], got["reasons"])
+        self.assertEqual(got["remote"], "prgs")
+        self.assertEqual(got["branch"], "master")
+        self.assertEqual(got["tracking_ref"], "refs/remotes/prgs/master")
+        self.assertEqual(got["repository_slug"], PRGS_SLUG)
+        self.assertEqual(got["source"], crr.BASE_REF_SOURCE_CONFIGURED_UPSTREAM)
+        # Cache and upstream agree here, which is the ordinary PRGS state.
+        self.assertFalse(got["cached_remote_head_conflicts"])
+        self.assertEqual(root_checkout_guard.resolve_remote_master_sha(self.root), head)
+
+    def test_prgs_resolves_without_a_configured_upstream(self):
+        """A PRGS checkout with no tracking config still resolves master."""
+        head = _make_repo(self.root, remote="prgs", url=PRGS_URL)
+        _set_remote_branch(self.root, "prgs", "master", head)
+
+        got = crr.resolve_target_base_ref(self.root)
+        self.assertTrue(got["proven"], got["reasons"])
+        self.assertEqual(got["tracking_ref"], "refs/remotes/prgs/master")
+        self.assertEqual(got["source"], crr.BASE_REF_SOURCE_UNIQUE_CANDIDATE)
+
+    def test_legacy_explicit_remote_refs_path_is_untouched(self):
+        """An explicit remote_refs override still short-circuits derivation."""
+        head = _make_repo(self.root, remote="prgs", url=PRGS_URL)
+        _set_remote_branch(self.root, "prgs", "master", head)
+
+        state = root_checkout_guard.resolve_remote_master_ref_state(
+            self.root, remote_refs=root_checkout_guard.REMOTE_MASTER_REFS
+        )
+        self.assertEqual(state["sha"], head)
+        self.assertEqual(state["source"], "explicit_remote_refs")
+
+
+class TestStaleCachedRemoteHead(_RepoCase):
+    """Required coverage 3: configured upstream MDCPS/dev vs stale cache -> main.
+
+    This is B1. The fixture deliberately does **not** point
+    ``refs/remotes/MDCPS/HEAD`` at ``dev``; it reproduces the disagreement that
+    was live on ``/Users/jasonwalker/Development/weekly-briefings``.
+    """
+
+    def test_configured_upstream_beats_stale_cached_remote_head(self):
+        head = self._weekly_briefings_shape()
+
+        # Precondition: the fixture really is in the defective state.
+        self.assertEqual(
+            _git(self.root, "symbolic-ref", "refs/remotes/MDCPS/HEAD"),
+            "refs/remotes/MDCPS/main",
+        )
+        self.assertEqual(_git(self.root, "config", "--get", "branch.dev.merge"), "refs/heads/dev")
+        self.assertNotEqual(
+            _git(self.root, "rev-parse", "refs/remotes/MDCPS/main"),
+            _git(self.root, "rev-parse", "refs/remotes/MDCPS/dev"),
+        )
+
+        got = crr.resolve_target_base_ref(self.root)
+        self.assertTrue(got["proven"], got["reasons"])
+        self.assertEqual(got["branch"], "dev")
+        self.assertEqual(got["tracking_ref"], "refs/remotes/MDCPS/dev")
+        self.assertEqual(got["source"], crr.BASE_REF_SOURCE_CONFIGURED_UPSTREAM)
+        # The stale cache is reported, never obeyed.
+        self.assertEqual(got["cached_remote_head_branch"], "main")
+        self.assertTrue(got["cached_remote_head_conflicts"])
+        self.assertEqual(root_checkout_guard.resolve_remote_master_sha(self.root), head)
+
+    def test_checkout_on_its_integration_tip_is_not_blocked(self):
+        """The live symptom: a checkout exactly on its tip was reported stale."""
+        head = self._weekly_briefings_shape()
+
+        state = root_checkout_guard.resolve_remote_master_ref_state(self.root)
+        self.assertEqual(state["sha"], head)
+        self.assertTrue(state["cached_remote_head_conflicts"])
+
+        assessment = root_checkout_guard.assess_root_checkout_guard(
+            workspace_path=self.root,
+            canonical_repo_root=self.root,
+            current_branch="dev",
+            head_sha=head,
+            porcelain_status="",
+            remote_master_sha=state["sha"],
+            remote_master_ref=state["ref"],
+        )
+        self.assertTrue(assessment["proven"], assessment["reasons"])
+
+        report = master_parity_gate.assess_target_repository_parity(
+            canonical_root=self.root, source="test"
+        )
+        self.assertFalse(report["stale"])
+        self.assertEqual(report["base_branch"], "dev")
+        self.assertEqual(report["reasons"], [])
+
+    def test_cached_remote_head_alone_never_proves_a_target(self):
+        """With no configured upstream, the cache cannot supply the branch."""
+        head = _make_repo(self.root, remote="MDCPS", url=MDCPS_URL)
+        # Only a non-candidate branch exists, and only the cache names it.
+        _set_remote_branch(self.root, "MDCPS", "trunk", head)
+        _set_remote_head(self.root, "MDCPS", "trunk")
+
+        got = crr.resolve_target_base_ref(self.root)
+        self.assertFalse(got["proven"])
+        self.assertEqual(got["reason_code"], crr.DENY_NO_BASE_BRANCH)
+        self.assertEqual(got["tracking_refs"], ())
+        self.assertEqual(got["cached_remote_head_branch"], "trunk")
+        # The refusal names the misleading signal so an operator is not sent
+        # chasing a ref that looks authoritative.
+        self.assertIn("trunk", " ".join(got["reasons"]))
+
+    def test_cached_remote_head_never_breaks_a_tie(self):
+        """Two candidates, no upstream: the cache must not decide."""
+        head = _make_repo(self.root, remote="MDCPS", url=MDCPS_URL)
+        _set_remote_branch(self.root, "MDCPS", "dev", head)
+        _set_remote_branch(self.root, "MDCPS", "main", head)
+        _set_remote_head(self.root, "MDCPS", "dev")
+
+        got = crr.resolve_target_base_ref(self.root)
+        self.assertFalse(got["proven"])
+        self.assertEqual(got["reason_code"], crr.DENY_AMBIGUOUS_BASE_BRANCH)
+        self.assertIsNone(root_checkout_guard.resolve_remote_master_sha(self.root))
+
+    def test_no_proven_source_is_the_remote_head_cache(self):
+        """Structural guard: the cache is not in the set of proving sources."""
+        sources = {
+            crr.BASE_REF_SOURCE_CONFIGURED_UPSTREAM,
+            crr.BASE_REF_SOURCE_UNIQUE_CANDIDATE,
+        }
+        self.assertNotIn("remote_head_symref", sources)
+        self.assertFalse(hasattr(crr, "BASE_REF_SOURCE_REMOTE_HEAD"))
+
+
+class TestCrossRepositoryTarget(_RepoCase):
+    """Required coverage 2/4/5: MDCPS/dev, no origin remote, exact case."""
+
+    def test_mdcps_dev_resolves(self):
+        head = self._weekly_briefings_shape()
+        got = crr.resolve_target_base_ref(self.root)
+        self.assertTrue(got["proven"], got["reasons"])
+        self.assertEqual(got["remote"], "MDCPS")
+        self.assertEqual(got["branch"], "dev")
+        self.assertEqual(got["tracking_ref"], "refs/remotes/MDCPS/dev")
+        self.assertEqual(got["repository_slug"], MDCPS_SLUG)
+        self.assertEqual(root_checkout_guard.resolve_remote_master_sha(self.root), head)
+
+    def test_no_remote_named_origin(self):
+        self._weekly_briefings_shape()
+        self.assertEqual(_git(self.root, "remote"), "MDCPS")
+        got = crr.resolve_target_base_ref(self.root)
+        self.assertTrue(got["proven"], got["reasons"])
+        self.assertNotIn("origin", got["tracking_ref"])
+
+    def test_remote_name_case_is_preserved_exactly(self):
+        self._weekly_briefings_shape()
+        got = crr.resolve_target_base_ref(self.root)
+        self.assertEqual(got["remote"], "MDCPS")
+        self.assertNotEqual(got["remote"], "mdcps")
+        # The tracking ref must address the real ref, which is case-sensitive.
+        self.assertEqual(got["tracking_ref"], "refs/remotes/MDCPS/dev")
+        self.assertTrue(
+            _git(self.root, "rev-parse", "--verify", got["tracking_ref"]),
+            "case-preserved tracking ref must resolve",
+        )
+
+    def test_lowercase_candidate_never_supplies_the_remote_name(self):
+        """Guards against silently case-folding MDCPS to the candidate 'mdcps'.
+
+        ``_IDENTITY_REMOTE_CANDIDATES`` contains a lowercase ``mdcps`` entry and
+        is probed *before* the repository's own remote listing. Git remote names
+        live in case-sensitive config subsections on every platform, so the
+        lowercase probe cannot resolve and the exact-case name must arrive from
+        ``git remote``. Asserted through config rather than ref lookup because a
+        case-insensitive filesystem (macOS) resolves loose refs either way, which
+        would make a ref-based assertion test the filesystem instead of the code.
+        """
+        self._weekly_briefings_shape()
+        res = subprocess.run(
+            ["git", "-C", self.root, "remote", "get-url", "mdcps"],
+            capture_output=True,
+            text=True,
+            check=False,
+        )
+        self.assertNotEqual(res.returncode, 0, "git remote names are case-sensitive")
+
+        # A caller naming the wrong case cannot disambiguate, but the target is
+        # unambiguous anyway, so the exact-case name is still resolved.
+        got = crr.resolve_target_base_ref(self.root, explicit_remote="mdcps")
+        self.assertTrue(got["proven"], got["reasons"])
+        self.assertEqual(got["remote"], "MDCPS")
+        self.assertEqual(got["tracking_ref"], "refs/remotes/MDCPS/dev")
+        self.assertFalse(got["identity_explicit"], "a non-matching name is not explicit intent")
+
+        name, slug = crr.resolve_identity_remote(self.root)
+        self.assertEqual(name, "MDCPS")
+        self.assertEqual(slug, MDCPS_SLUG)
+
+
+class TestTargetStaleness(_RepoCase):
+    """Required coverage 6/7: matching tip, and behind or divergent checkout."""
+
+    def test_local_equal_to_resolved_tip_is_not_stale(self):
+        self._weekly_briefings_shape()
+        got = master_parity_gate.assess_target_repository_parity(
+            canonical_root=self.root, source="test"
+        )
+        self.assertTrue(got["determinable"])
+        self.assertFalse(got["stale"])
+        self.assertEqual(got["tracking_ref"], "refs/remotes/MDCPS/dev")
+        self.assertEqual(got["base_remote"], "MDCPS")
+        self.assertEqual(got["base_branch"], "dev")
+        self.assertIsNone(got["reason_code"])
+
+    def test_local_behind_resolved_tip_is_stale(self):
+        head = self._weekly_briefings_shape()
+        advanced = _advance(self.root, "remote moved on")
+        _set_remote_branch(self.root, "MDCPS", "dev", advanced)
+        _git(self.root, "reset", "--hard", "--quiet", head)
+
+        got = master_parity_gate.assess_target_repository_parity(
+            canonical_root=self.root, source="test"
+        )
+        self.assertTrue(got["determinable"])
+        self.assertTrue(got["stale"])
+        self.assertEqual(got["checkout_head"], head)
+        self.assertEqual(got["remote_tracking_head"], advanced)
+
+    def test_local_divergent_from_resolved_tip_is_stale(self):
+        head = self._weekly_briefings_shape()
+        remote_side = _advance(self.root, "remote side")
+        _set_remote_branch(self.root, "MDCPS", "dev", remote_side)
+        _git(self.root, "reset", "--hard", "--quiet", head)
+        local_side = _advance(self.root, "local side")
+
+        got = master_parity_gate.assess_target_repository_parity(
+            canonical_root=self.root, source="test"
+        )
+        self.assertTrue(got["stale"])
+        self.assertEqual(got["checkout_head"], local_side)
+        self.assertNotEqual(local_side, remote_side)
+
+
+class TestFailClosed(_RepoCase):
+    """Required coverage 8/9: missing remote or ref, and ambiguous targets."""
+
+    def test_missing_remote_fails_closed(self):
+        _make_repo(self.root, remote=None, url=None)
+        got = crr.resolve_target_base_ref(self.root)
+        self.assertFalse(got["proven"])
+        self.assertEqual(got["reason_code"], crr.DENY_NO_IDENTITY_REMOTE)
+        self.assertEqual(got["tracking_refs"], ())
+        self.assertIsNone(root_checkout_guard.resolve_remote_master_sha(self.root))
+
+    def test_missing_tracking_ref_fails_closed(self):
+        _make_repo(self.root, remote="MDCPS", url=MDCPS_URL)
+        # Remote configured, but nothing has ever been fetched.
+        got = crr.resolve_target_base_ref(self.root)
+        self.assertFalse(got["proven"])
+        self.assertEqual(got["reason_code"], crr.DENY_NO_BASE_BRANCH)
+        self.assertIsNone(root_checkout_guard.resolve_remote_master_sha(self.root))
+
+    def test_configured_upstream_without_a_tracking_ref_fails_closed(self):
+        """A proven target always names a ref that resolves."""
+        _make_repo(self.root, remote="MDCPS", url=MDCPS_URL)
+        _set_upstream(self.root, "MDCPS", "dev")  # declared, never fetched
+
+        got = crr.resolve_target_base_ref(self.root)
+        self.assertFalse(got["proven"])
+        self.assertEqual(got["reason_code"], crr.DENY_NO_BASE_BRANCH)
+        self.assertIsNone(got["tracking_ref"])
+
+    def test_every_proven_target_resolves(self):
+        """No 'proven' result may name an unresolvable tracking ref."""
+        head = self._weekly_briefings_shape()
+        got = crr.resolve_target_base_ref(self.root)
+        self.assertTrue(got["proven"])
+        self.assertEqual(_git(self.root, "rev-parse", got["tracking_ref"]), head)
+
+    def test_ambiguous_integration_branch_fails_closed(self):
+        head = _make_repo(self.root, remote="MDCPS", url=MDCPS_URL)
+        # Two candidate integration branches and no configured upstream.
+        _set_remote_branch(self.root, "MDCPS", "dev", head)
+        _set_remote_branch(self.root, "MDCPS", "main", head)
+
+        got = crr.resolve_target_base_ref(self.root)
+        self.assertFalse(got["proven"])
+        self.assertEqual(got["reason_code"], crr.DENY_AMBIGUOUS_BASE_BRANCH)
+        self.assertIsNone(root_checkout_guard.resolve_remote_master_sha(self.root))
+
+    def test_ambiguous_identity_remote_fails_closed(self):
+        head = _make_repo(self.root, remote="MDCPS", url=MDCPS_URL)
+        _git(self.root, "remote", "add", "prgs", PRGS_URL)
+        _set_remote_branch(self.root, "MDCPS", "dev", head)
+        _set_remote_branch(self.root, "prgs", "master", head)
+
+        got = crr.resolve_target_base_ref(self.root)
+        self.assertFalse(got["proven"])
+        self.assertEqual(got["reason_code"], crr.DENY_AMBIGUOUS_REMOTE)
+        self.assertEqual(got["tracking_refs"], ())
+
+    def test_orphan_tracking_ref_from_removed_remote_is_ignored(self):
+        """The live Gitea-Tools symptom: refs/remotes/origin/* outlives its remote."""
+        head = _make_repo(self.root, remote="prgs", url=PRGS_URL)
+        _set_remote_branch(self.root, "prgs", "master", head)
+        _set_upstream(self.root, "prgs", "master")
+        # An abandoned ref left behind by a remote that no longer exists.
+        _set_remote_branch(self.root, "origin", "master", head)
+        _advance(self.root, "orphan must not be consulted")
+
+        got = master_parity_gate.assess_target_repository_parity(
+            canonical_root=self.root, source="test"
+        )
+        self.assertEqual(got["tracking_ref"], "refs/remotes/prgs/master")
+        self.assertEqual(got["repository_slug"], PRGS_SLUG)
+        self.assertNotIn(
+            "target repository identity could not be derived from its git remote",
+            got["reasons"],
+        )
+
+
+class TestExplicitVersusInferredRemote(_RepoCase):
+    """Required coverage 11: explicit disambiguation stays distinct from inference.
+
+    This is B2. A remote the module inferred while probing must never re-enter
+    the resolver as though an operator had named it.
+    """
+
+    def _two_remotes(self) -> str:
+        head = _make_repo(self.root, remote="MDCPS", url=MDCPS_URL)
+        _git(self.root, "remote", "add", "prgs", PRGS_URL)
+        _set_remote_branch(self.root, "MDCPS", "dev", head)
+        _set_remote_branch(self.root, "prgs", "master", head)
+        return head
+
+    def test_explicit_remote_disambiguates(self):
+        self._two_remotes()
+        got = crr.resolve_target_base_ref(self.root, explicit_remote="MDCPS")
+        self.assertTrue(got["proven"], got["reasons"])
+        self.assertEqual(got["remote"], "MDCPS")
+        self.assertEqual(got["branch"], "dev")
+        self.assertTrue(got["identity_explicit"])
+
+    def test_inferred_remote_does_not_disambiguate(self):
+        """Feeding the inferred remote back in must not unlock the target."""
+        self._two_remotes()
+        inferred, _ = crr.resolve_identity_remote(self.root)
+        self.assertIsNotNone(inferred, "the first-wins probe still returns a name")
+
+        # The report infers internally and must still refuse.
+        report = master_parity_gate.assess_target_repository_parity(
+            canonical_root=self.root, source="test"
+        )
+        self.assertEqual(report["reason_code"], crr.DENY_AMBIGUOUS_REMOTE)
+        self.assertIsNone(report["repository_slug"])
+        self.assertIsNone(report["tracking_ref"])
+        self.assertFalse(report["stale"])
+        self.assertTrue(report["reasons"])
+
+    def test_report_never_passes_a_remote_into_the_resolver(self):
+        """Structural guard against the exact B2 regression."""
+        src = inspect.getsource(master_parity_gate.assess_target_repository_parity)
+        self.assertIn("resolve_target_base_ref(canonical_root)", src)
+        self.assertNotIn("resolve_target_base_ref(canonical_root, remote=", src)
+        self.assertNotIn("explicit_remote=identity", src)
+        # Reporting must use the ambiguity-aware identity resolver.
+        self.assertIn("assess_identity_remote(canonical_root)", src)
+
+    def test_explicit_parameter_is_named_for_its_meaning(self):
+        for fn in (
+            crr.resolve_target_base_ref,
+            root_checkout_guard.resolve_remote_master_sha,
+            root_checkout_guard.resolve_remote_master_ref_state,
+        ):
+            params = inspect.signature(fn).parameters
+            self.assertIn("explicit_remote", params, fn.__name__)
+            self.assertNotIn("remote", params, fn.__name__)
+
+    def test_unmatched_explicit_remote_cannot_unlock_an_ambiguous_target(self):
+        self._two_remotes()
+        got = crr.resolve_target_base_ref(self.root, explicit_remote="nonexistent")
+        self.assertFalse(got["proven"])
+        self.assertEqual(got["reason_code"], crr.DENY_AMBIGUOUS_REMOTE)
+
+
+class TestGatingAndReportingAgree(_RepoCase):
+    """Required coverage 10: guard and parity report make identical decisions."""
+
+    def test_same_resolved_target_for_gate_and_report(self):
+        self._weekly_briefings_shape()
+
+        gate = root_checkout_guard.resolve_remote_master_ref_state(self.root)
+        report = master_parity_gate.assess_target_repository_parity(
+            canonical_root=self.root, source="test"
+        )
+
+        self.assertEqual(gate["remote"], report["base_remote"])
+        self.assertEqual(gate["branch"], report["base_branch"])
+        self.assertEqual(gate["sha"], report["remote_tracking_head"])
+        self.assertIn(
+            gate["ref"],
+            (report["tracking_ref"], f"{gate['remote']}/{gate['branch']}"),
+        )
+
+    def test_both_sides_refuse_the_same_unresolvable_target(self):
+        _make_repo(self.root, remote=None, url=None)
+
+        gate = root_checkout_guard.resolve_remote_master_ref_state(self.root)
+        report = master_parity_gate.assess_target_repository_parity(
+            canonical_root=self.root, source="test"
+        )
+        self.assertIsNone(gate["sha"])
+        self.assertIsNone(report["remote_tracking_head"])
+        self.assertFalse(report["stale"])
+        self.assertTrue(report["reasons"])
+        self.assertEqual(gate["reason_code"], report["reason_code"])
+
+    def test_both_sides_refuse_the_same_ambiguous_target(self):
+        head = _make_repo(self.root, remote="MDCPS", url=MDCPS_URL)
+        _git(self.root, "remote", "add", "prgs", PRGS_URL)
+        _set_remote_branch(self.root, "MDCPS", "dev", head)
+        _set_remote_branch(self.root, "prgs", "master", head)
+
+        gate = root_checkout_guard.resolve_remote_master_ref_state(self.root)
+        report = master_parity_gate.assess_target_repository_parity(
+            canonical_root=self.root, source="test"
+        )
+
+        self.assertIsNone(gate["sha"])
+        self.assertEqual(gate["reason_code"], crr.DENY_AMBIGUOUS_REMOTE)
+        # The report must not name a repository the gate refuses to act on.
+        self.assertEqual(report["reason_code"], gate["reason_code"])
+        self.assertIsNone(report["repository_slug"])
+        self.assertIsNone(report["base_remote"])
+        self.assertIsNone(report["base_branch"])
+        self.assertFalse(report["stale"])
+
+    def test_both_sides_refuse_the_same_ambiguous_branch(self):
+        head = _make_repo(self.root, remote="MDCPS", url=MDCPS_URL)
+        _set_remote_branch(self.root, "MDCPS", "dev", head)
+        _set_remote_branch(self.root, "MDCPS", "main", head)
+
+        gate = root_checkout_guard.resolve_remote_master_ref_state(self.root)
+        report = master_parity_gate.assess_target_repository_parity(
+            canonical_root=self.root, source="test"
+        )
+        self.assertEqual(gate["reason_code"], crr.DENY_AMBIGUOUS_BASE_BRANCH)
+        self.assertEqual(report["reason_code"], gate["reason_code"])
+        self.assertIsNone(report["tracking_ref"])
+
+
+class TestProductionCallers(unittest.TestCase):
+    """Required coverage 13: every production caller consumes the same target."""
+
+    def test_guard_reports_the_ref_it_actually_compared(self):
+        assessment = root_checkout_guard.assess_root_checkout_guard(
+            workspace_path="/tmp/nonexistent-workspace-983",
+            canonical_repo_root="/tmp/nonexistent-root-983",
+            current_branch="dev",
+            head_sha="a" * 40,
+            porcelain_status="",
+            remote_master_sha="b" * 40,
+            remote_master_ref="refs/remotes/MDCPS/dev",
+        )
+        self.assertTrue(assessment["block"])
+        joined = " ".join(assessment["reasons"])
+        self.assertIn("refs/remotes/MDCPS/dev", joined)
+        self.assertNotIn("prgs/master", joined)
+
+    def test_guard_message_without_a_ref_stays_generic(self):
+        assessment = root_checkout_guard.assess_root_checkout_guard(
+            workspace_path="/tmp/nonexistent-workspace-983",
+            canonical_repo_root="/tmp/nonexistent-root-983",
+            current_branch="master",
+            head_sha="a" * 40,
+            porcelain_status="",
+            remote_master_sha="b" * 40,
+        )
+        joined = " ".join(assessment["reasons"])
+        self.assertIn("the tracking integration ref", joined)
+        self.assertNotIn("prgs/master", joined)
+
+    def test_anti_stomp_preflight_forwards_the_resolved_ref(self):
+        sig = inspect.signature(anti_stomp_preflight.assess_anti_stomp_preflight)
+        self.assertIn("remote_master_ref", sig.parameters)
+        src = inspect.getsource(anti_stomp_preflight.assess_anti_stomp_preflight)
+        self.assertIn("remote_master_ref=remote_master_ref", src)
+
+    def test_no_production_caller_inherits_the_prgs_default(self):
+        """Every resolve site must derive, or pass remote_refs explicitly."""
+        import gitea_mcp_server
+
+        src = inspect.getsource(gitea_mcp_server)
+        # The four historical call sites now consume the resolved-target state.
+        self.assertGreaterEqual(src.count("resolve_remote_master_ref_state("), 4)
+        self.assertNotIn("resolve_remote_master_sha(canonical_root)", src)
+
+    def test_no_production_caller_supplies_an_explicit_remote(self):
+        """Nothing in production may suppress the ambiguity gate (#983 B2)."""
+        import gitea_mcp_server
+
+        for module in (gitea_mcp_server, anti_stomp_preflight, master_parity_gate):
+            src = inspect.getsource(module)
+            self.assertNotIn("explicit_remote=", src, module.__name__)
+
+
+class TestRepositoryStructureUntouched(_RepoCase):
+    """Required coverage 12: resolution mutates no ref, remote, config, or checkout."""
+
+    def test_resolution_creates_no_refs_or_branches(self):
+        self._weekly_briefings_shape()
+
+        fmt = "--format=%(refname) %(objectname)"
+        before_refs = _git(self.root, "for-each-ref", fmt)
+        before_remotes = _git(self.root, "remote")
+        before_config = _git(self.root, "config", "--local", "--list")
+        before_head = _git(self.root, "rev-parse", "HEAD")
+        before_branch = _git(self.root, "symbolic-ref", "--short", "HEAD")
+        before_status = _git(self.root, "status", "--porcelain", "--untracked-files=all")
+        before_symref = _git(self.root, "symbolic-ref", "refs/remotes/MDCPS/HEAD")
+
+        crr.resolve_target_base_ref(self.root)
+        crr.assess_identity_remote(self.root)
+        root_checkout_guard.resolve_remote_master_sha(self.root)
+        root_checkout_guard.resolve_remote_master_ref_state(self.root)
+        master_parity_gate.assess_target_repository_parity(
+            canonical_root=self.root, source="test"
+        )
+
+        self.assertEqual(_git(self.root, "for-each-ref", fmt), before_refs)
+        self.assertEqual(_git(self.root, "remote"), before_remotes)
+        self.assertEqual(_git(self.root, "config", "--local", "--list"), before_config)
+        self.assertEqual(_git(self.root, "rev-parse", "HEAD"), before_head)
+        self.assertEqual(_git(self.root, "symbolic-ref", "--short", "HEAD"), before_branch)
+        self.assertEqual(
+            _git(self.root, "status", "--porcelain", "--untracked-files=all"), before_status
+        )
+        # The stale cache is specifically NOT repaired: that would be a mutation.
+        self.assertEqual(_git(self.root, "symbolic-ref", "refs/remotes/MDCPS/HEAD"), before_symref)
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_issue_985_project_scoped_launcher.py

@@ -0,0 +1,664 @@
+"""#985 — trusted client-instance identity for project-scoped LLM launches.
+
+Covers the acceptance criteria of issue #985:
+
+* a three-namespace (author/reviewer/merger) launch succeeds with no
+  controller or reconciler profile,
+* every worker of one launch shares exactly one trusted instance ID,
+* two concurrent launches receive distinct IDs and stay distinguishable,
+* reuse, missing, legacy, malformed, unsanctioned, and unsealed identities all
+  fail closed,
+* the existing five-namespace behaviour is unchanged,
+* Claude Code has a documented, runnable launch command.
+"""
+
+from __future__ import annotations
+
+import datetime as _dt
+import json
+import os
+import sys
+import unittest
+
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+import mcp_application_launcher as launcher  # noqa: E402
+import mcp_fleet_snapshot as fleet  # noqa: E402
+
+NOW = _dt.datetime(2026, 7, 31, 12, 0, 0, tzinfo=_dt.timezone.utc)
+
+PROJECT_NAMESPACES = ("author", "reviewer", "merger")
+ALL_NAMESPACES = ("author", "reviewer", "merger", "controller", "reconciler")
+
+#: A project-scoped configuration intentionally has no controller/reconciler
+#: profile — that absence is the condition under test, not an omission.
+PROJECT_PROFILES = {ns: f"prgs-{ns}" for ns in PROJECT_NAMESPACES}
+FULL_PROFILES = {ns: f"prgs-{ns}" for ns in ALL_NAMESPACES}
+
+
+def _instance_ids(servers):
+    return {
+        name: entry["env"][launcher.CLIENT_INSTANCE_ENV]
+        for name, entry in servers.items()
+    }
+
+
+class ResolveLaunchNamespacesTests(unittest.TestCase):
+    """The validated subset allow-list."""
+
+    def test_none_resolves_to_all_five_in_canonical_order(self):
+        self.assertEqual(
+            launcher.resolve_launch_namespaces(None), tuple(ALL_NAMESPACES)
+        )
+
+    def test_project_subset_resolves_in_canonical_order(self):
+        # Deliberately out of order on input; canonical order on output.
+        self.assertEqual(
+            launcher.resolve_launch_namespaces(["merger", "author", "reviewer"]),
+            ("author", "reviewer", "merger"),
+        )
+
+    def test_case_and_whitespace_normalised(self):
+        self.assertEqual(
+            launcher.resolve_launch_namespaces([" Author ", "REVIEWER"]),
+            ("author", "reviewer"),
+        )
+
+    def test_unsanctioned_namespace_refused(self):
+        with self.assertRaises(ValueError) as ctx:
+            launcher.resolve_launch_namespaces(["author", "admin"])
+        self.assertIn("admin", str(ctx.exception))
+
+    def test_typo_is_refused_not_silently_dropped(self):
+        with self.assertRaises(ValueError):
+            launcher.resolve_launch_namespaces(["author", "reviewr"])
+
+    def test_empty_subset_refused(self):
+        with self.assertRaises(ValueError):
+            launcher.resolve_launch_namespaces([])
+        with self.assertRaises(ValueError):
+            launcher.resolve_launch_namespaces(["   "])
+
+    def test_duplicate_namespace_refused(self):
+        with self.assertRaises(ValueError):
+            launcher.resolve_launch_namespaces(["author", "author"])
+
+
+class ProjectScopedLaunchTests(unittest.TestCase):
+    """AC: three-namespace launch without controller/reconciler profiles."""
+
+    def _build(self, **kwargs):
+        kwargs.setdefault("client_type", "claude_code")
+        kwargs.setdefault("namespaces", PROJECT_NAMESPACES)
+        kwargs.setdefault("command", "python3")
+        kwargs.setdefault("args", ["mcp_server.py"])
+        profiles = kwargs.pop("profiles", PROJECT_PROFILES)
+        return launcher.build_application_mcp_servers(profiles, **kwargs)
+
+    def test_three_namespace_launch_succeeds_without_controller_reconciler(self):
+        built = self._build(launch_nonce="proj-1", now=NOW)
+        servers = built["mcpServers"]
+        self.assertEqual(
+            sorted(servers), ["gitea-author", "gitea-merger", "gitea-reviewer"]
+        )
+        self.assertEqual(built["namespace_count"], 3)
+        self.assertEqual(list(built["namespaces"]), list(PROJECT_NAMESPACES))
+
+    def test_excluded_namespaces_are_not_started(self):
+        built = self._build(launch_nonce="proj-2", now=NOW)
+        self.assertNotIn("gitea-controller", built["mcpServers"])
+        self.assertNotIn("gitea-reconciler", built["mcpServers"])
+        self.assertEqual(
+            sorted(built["excluded_namespaces"]), ["controller", "reconciler"]
+        )
+        self.assertTrue(built["project_scoped"])
+
+    def test_all_workers_of_one_launch_share_one_trusted_id(self):
+        built = self._build(launch_nonce="proj-3", now=NOW)
+        ids = set(_instance_ids(built["mcpServers"]).values())
+        self.assertEqual(len(ids), 1, ids)
+        shared = next(iter(ids))
+        self.assertTrue(fleet.assess_instance_identity(shared)["trusted"])
+        self.assertEqual(built["client_instance_id"], shared)
+
+    def test_every_worker_env_is_launcher_sealed(self):
+        built = self._build(launch_nonce="proj-4", now=NOW)
+        for name, entry in built["mcpServers"].items():
+            env = entry["env"]
+            self.assertEqual(
+                env[launcher.INSTANCE_PROVENANCE_ENV],
+                fleet.INSTANCE_ID_PROVENANCE_TRUSTED,
+                name,
+            )
+            self.assertEqual(env["GITEA_CLIENT_MANAGED"], "1", name)
+            self.assertFalse(
+                env[launcher.CLIENT_INSTANCE_ENV].startswith("legacy-"), name
+            )
+
+    def test_attribution_proof_holds_for_three_namespace_launch(self):
+        """A subset launch must not read as 'missing two workers'."""
+        built = self._build(launch_nonce="proj-5", now=NOW)
+        proof = launcher.collect_instance_ids_from_mcp_servers(
+            built["mcpServers"], namespaces=PROJECT_NAMESPACES
+        )
+        self.assertTrue(proof["shared_single_trusted_id"], proof)
+        self.assertEqual(proof["missing_servers"], [])
+        # And without an explicit expectation it inspects what is present.
+        inferred = launcher.collect_instance_ids_from_mcp_servers(
+            built["mcpServers"]
+        )
+        self.assertTrue(inferred["shared_single_trusted_id"], inferred)
+
+    def test_missing_profile_for_requested_namespace_refused(self):
+        with self.assertRaises(ValueError) as ctx:
+            self._build(profiles={"author": "prgs-author"})
+        self.assertIn("merger", str(ctx.exception))
+
+    def test_profile_for_unlaunched_namespace_is_ignored_not_started(self):
+        """Extra profiles never widen the launch beyond the validated subset."""
+        built = self._build(profiles=FULL_PROFILES, launch_nonce="proj-6", now=NOW)
+        self.assertEqual(len(built["mcpServers"]), 3)
+        self.assertNotIn("gitea-controller", built["mcpServers"])
+
+    def test_unsanctioned_namespace_refused_at_build(self):
+        with self.assertRaises(ValueError):
+            self._build(
+                profiles={**PROJECT_PROFILES, "admin": "prgs-admin"},
+                namespaces=["author", "admin"],
+            )
+
+
+class ConcurrentLaunchDistinctionTests(unittest.TestCase):
+    """AC: concurrent launches are distinct and distinguishable."""
+
+    def _build(self, nonce, namespaces=PROJECT_NAMESPACES, profiles=None):
+        return launcher.build_application_mcp_servers(
+            profiles or PROJECT_PROFILES,
+            client_type="claude_code",
+            namespaces=namespaces,
+            launch_nonce=nonce,
+            now=NOW,
+            command="python3",
+            args=["mcp_server.py"],
+        )
+
+    def test_two_concurrent_launches_receive_different_ids(self):
+        a = self._build("concurrent-a")
+        b = self._build("concurrent-b")
+        self.assertNotEqual(a["client_instance_id"], b["client_instance_id"])
+
+    def test_same_profile_across_launches_stays_distinguishable(self):
+        """Same profile, two launches: distinguishable, so not a false duplicate."""
+        a = self._build("dist-a")
+        b = self._build("dist-b")
+        a_author = a["mcpServers"]["gitea-author"]["env"]
+        b_author = b["mcpServers"]["gitea-author"]["env"]
+        # Same profile on purpose — that alone must never imply duplication.
+        self.assertEqual(a_author["GITEA_MCP_PROFILE"], b_author["GITEA_MCP_PROFILE"])
+        self.assertNotEqual(
+            a_author[launcher.CLIENT_INSTANCE_ENV],
+            b_author[launcher.CLIENT_INSTANCE_ENV],
+        )
+        for env in (a_author, b_author):
+            assessed = launcher.inherit_or_refuse_client_instance(env)
+            self.assertTrue(assessed["trusted"], assessed)
+            self.assertTrue(assessed["launcher_sealed"], assessed)
+
+    def test_no_nonce_still_yields_distinct_ids(self):
+        a = self._build(None)
+        b = self._build(None)
+        self.assertNotEqual(a["client_instance_id"], b["client_instance_id"])
+
+    def test_ids_are_never_persisted_between_launches(self):
+        """No static per-project ID: two runs of one config differ."""
+        first = self._build("static-check-1")
+        second = self._build("static-check-2")
+        self.assertNotEqual(
+            first["mcpServers"]["gitea-author"]["env"][launcher.CLIENT_INSTANCE_ENV],
+            second["mcpServers"]["gitea-author"]["env"][launcher.CLIENT_INSTANCE_ENV],
+        )
+
+
+class FailClosedIdentityTests(unittest.TestCase):
+    """AC: missing / legacy / reused / conflicting / unsealed all fail closed."""
+
+    def test_missing_identity_fails_closed(self):
+        assessed = launcher.inherit_or_refuse_client_instance({})
+        self.assertFalse(assessed["trusted"])
+        self.assertFalse(assessed["launcher_sealed"])
+        self.assertEqual(
+            assessed["provenance"], fleet.INSTANCE_ID_PROVENANCE_MISSING
+        )
+
+    def test_legacy_pid_identity_fails_closed(self):
+        for legacy in ("legacy-pid-40578", "pid-1234", "proc-99"):
+            with self.subTest(legacy=legacy):
+                assessed = launcher.inherit_or_refuse_client_instance(
+                    {
+                        launcher.CLIENT_INSTANCE_ENV: legacy,
+                        launcher.INSTANCE_PROVENANCE_ENV: (
+                            fleet.INSTANCE_ID_PROVENANCE_TRUSTED
+                        ),
+                    }
+                )
+                self.assertFalse(assessed["trusted"], legacy)
+                self.assertFalse(assessed["launcher_sealed"], legacy)
+
+    def test_wellformed_but_unsealed_identity_fails_closed(self):
+        """The decisive case: format alone must not manufacture trust."""
+        forged = fleet.generate_client_instance_id(
+            "claude_code", launch_nonce="forged", now=NOW
+        )
+        # A hand-set env var with a valid-looking ID but no launcher seal.
+        assessed = launcher.inherit_or_refuse_client_instance(
+            {launcher.CLIENT_INSTANCE_ENV: forged}
+        )
+        self.assertFalse(assessed["trusted"], assessed)
+        self.assertFalse(assessed["launcher_sealed"], assessed)
+        self.assertEqual(
+            assessed["provenance"], fleet.INSTANCE_ID_PROVENANCE_UNSEALED
+        )
+        # The value is still reported for diagnosis, not silently dropped.
+        self.assertEqual(assessed["client_instance_id"], forged)
+
+    def test_wrong_provenance_marker_fails_closed(self):
+        forged = fleet.generate_client_instance_id(
+            "claude_code", launch_nonce="wrong-marker", now=NOW
+        )
+        assessed = launcher.inherit_or_refuse_client_instance(
+            {
+                launcher.CLIENT_INSTANCE_ENV: forged,
+                launcher.INSTANCE_PROVENANCE_ENV: "totally_trusted",
+            }
+        )
+        self.assertFalse(assessed["trusted"], assessed)
+        self.assertEqual(
+            assessed["provenance"], fleet.INSTANCE_ID_PROVENANCE_UNSEALED
+        )
+
+    def test_sealed_launcher_env_is_trusted(self):
+        env = launcher.namespace_worker_env(
+            profile_name="prgs-author",
+            client_type="claude_code",
+            client_instance_id=fleet.generate_client_instance_id(
+                "claude_code", launch_nonce="sealed", now=NOW
+            ),
+        )
+        assessed = launcher.inherit_or_refuse_client_instance(env)
+        self.assertTrue(assessed["trusted"], assessed)
+        self.assertTrue(assessed["launcher_sealed"], assessed)
+
+    def test_untrusted_id_refused_on_build_and_worker_env(self):
+        with self.assertRaises(ValueError):
+            launcher.build_application_mcp_servers(
+                PROJECT_PROFILES,
+                client_type="claude_code",
+                namespaces=PROJECT_NAMESPACES,
+                client_instance_id="legacy-pid-1",
+                command="python3",
+                args=["mcp_server.py"],
+            )
+        with self.assertRaises(ValueError):
+            launcher.namespace_worker_env(
+                profile_name="prgs-author",
+                client_type="claude_code",
+                client_instance_id="hand-written",
+            )
+
+    def test_extra_env_cannot_forge_seal_or_identity(self):
+        trusted = fleet.generate_client_instance_id(
+            "claude_code", launch_nonce="seal-guard", now=NOW
+        )
+        env = launcher.namespace_worker_env(
+            profile_name="prgs-author",
+            client_type="claude_code",
+            client_instance_id=trusted,
+            extra_env={
+                launcher.CLIENT_INSTANCE_ENV: "inst-spoof-20260101T000000Z-abcdef",
+                launcher.INSTANCE_PROVENANCE_ENV: "trusted_launcher",
+                "GITEA_CLIENT_MANAGED": "0",
+            },
+        )
+        self.assertEqual(env[launcher.CLIENT_INSTANCE_ENV], trusted)
+        self.assertEqual(env["GITEA_CLIENT_MANAGED"], "1")
+
+    def test_conflicting_ids_across_workers_break_shared_attribution(self):
+        """One launch whose workers disagree must not read as shared."""
+        built = launcher.build_application_mcp_servers(
+            PROJECT_PROFILES,
+            client_type="claude_code",
+            namespaces=PROJECT_NAMESPACES,
+            launch_nonce="conflict",
+            now=NOW,
+            command="python3",
+            args=["mcp_server.py"],
+        )
+        servers = built["mcpServers"]
+        servers["gitea-merger"]["env"][launcher.CLIENT_INSTANCE_ENV] = (
+            fleet.generate_client_instance_id(
+                "claude_code", launch_nonce="other", now=NOW
+            )
+        )
+        proof = launcher.collect_instance_ids_from_mcp_servers(
+            servers, namespaces=PROJECT_NAMESPACES
+        )
+        self.assertFalse(proof["shared_single_trusted_id"], proof)
+        self.assertEqual(len(proof["unique_instance_ids"]), 2)
+
+    def test_absent_expected_worker_breaks_shared_attribution(self):
+        built = launcher.build_application_mcp_servers(
+            PROJECT_PROFILES,
+            client_type="claude_code",
+            namespaces=PROJECT_NAMESPACES,
+            launch_nonce="absent",
+            now=NOW,
+            command="python3",
+            args=["mcp_server.py"],
+        )
+        servers = dict(built["mcpServers"])
+        servers.pop("gitea-merger")
+        proof = launcher.collect_instance_ids_from_mcp_servers(
+            servers, namespaces=PROJECT_NAMESPACES
+        )
+        self.assertFalse(proof["shared_single_trusted_id"], proof)
+        self.assertEqual(proof["missing_servers"], ["gitea-merger"])
+
+    def test_reused_trusted_id_is_reuse_not_a_fresh_launch(self):
+        """Resume reuses by design; two *live* launches sharing it is the hazard."""
+        first = launcher.build_application_mcp_servers(
+            PROJECT_PROFILES,
+            client_type="claude_code",
+            namespaces=PROJECT_NAMESPACES,
+            launch_nonce="reuse-src",
+            now=NOW,
+            command="python3",
+            args=["mcp_server.py"],
+        )
+        resumed = launcher.build_application_mcp_servers(
+            PROJECT_PROFILES,
+            client_type="claude_code",
+            namespaces=PROJECT_NAMESPACES,
+            client_instance_id=first["client_instance_id"],
+            command="python3",
+            args=["mcp_server.py"],
+        )
+        self.assertEqual(
+            first["client_instance_id"], resumed["client_instance_id"]
+        )
+        # Two independent launches must never collide by default.
+        independent = launcher.build_application_mcp_servers(
+            PROJECT_PROFILES,
+            client_type="claude_code",
+            namespaces=PROJECT_NAMESPACES,
+            command="python3",
+            args=["mcp_server.py"],
+        )
+        self.assertNotEqual(
+            first["client_instance_id"], independent["client_instance_id"]
+        )
+
+
+class BackwardCompatibilityTests(unittest.TestCase):
+    """AC: existing full five-namespace behaviour remains compatible."""
+
+    def test_omitting_namespaces_launches_all_five(self):
+        built = launcher.build_application_mcp_servers(
+            FULL_PROFILES,
+            client_type="codex",
+            launch_nonce="compat",
+            now=NOW,
+            command="python3",
+            args=["mcp_server.py"],
+        )
+        self.assertEqual(len(built["mcpServers"]), 5)
+        self.assertEqual(built["namespace_count"], 5)
+        self.assertEqual(built["excluded_namespaces"], [])
+        self.assertFalse(built["project_scoped"])
+        ids = set(_instance_ids(built["mcpServers"]).values())
+        self.assertEqual(len(ids), 1)
+
+    def test_five_namespace_missing_profile_still_refused(self):
+        incomplete = {ns: f"prgs-{ns}" for ns in PROJECT_NAMESPACES}
+        with self.assertRaises(ValueError) as ctx:
+            launcher.build_application_mcp_servers(
+                incomplete,
+                client_type="codex",
+                command="python3",
+                args=["mcp_server.py"],
+            )
+        message = str(ctx.exception)
+        self.assertIn("controller", message)
+        self.assertIn("reconciler", message)
+
+    def test_legacy_collect_call_without_namespaces_still_works(self):
+        built = launcher.build_application_mcp_servers(
+            FULL_PROFILES,
+            client_type="codex",
+            launch_nonce="compat-proof",
+            now=NOW,
+            command="python3",
+            args=["mcp_server.py"],
+        )
+        proof = launcher.collect_instance_ids_from_mcp_servers(built["mcpServers"])
+        self.assertTrue(proof["shared_single_trusted_id"], proof)
+        self.assertEqual(proof["namespace_server_count"], 5)
+
+    def test_gitea_config_wrapper_supports_subset(self):
+        import gitea_config
+
+        built = gitea_config.multi_namespace_launcher_entries(
+            PROJECT_PROFILES,
+            client_type="claude_code",
+            namespaces=PROJECT_NAMESPACES,
+            config_path="/cfg/profiles.json",
+            launch_nonce="cfg-subset",
+        )
+        self.assertEqual(built["namespace_count"], 3)
+        self.assertEqual(
+            sorted(built["excluded_namespaces"]), ["controller", "reconciler"]
+        )
+
+    def test_gitea_config_wrapper_default_is_five(self):
+        import gitea_config
+
+        built = gitea_config.multi_namespace_launcher_entries(
+            FULL_PROFILES,
+            client_type="claude_code",
+            config_path="/cfg/profiles.json",
+            launch_nonce="cfg-full",
+        )
+        self.assertEqual(built["namespace_count"], 5)
+
+
+class RunnableLaunchPathTests(unittest.TestCase):
+    """AC: Claude Code has a documented supported launch command."""
+
+    def test_claude_code_is_a_supported_client(self):
+        self.assertIn("claude_code", launcher.supported_launch_clients())
+
+    def test_claude_code_argv_uses_per_launch_config(self):
+        argv = launcher.build_client_launch_argv("claude_code", "/tmp/launch.json")
+        self.assertEqual(argv[0], "claude")
+        self.assertIn("--mcp-config", argv)
+        self.assertIn("/tmp/launch.json", argv)
+        self.assertIn("--strict-mcp-config", argv)
+
+    def test_extra_args_are_forwarded(self):
+        argv = launcher.build_client_launch_argv(
+            "claude_code", "/tmp/launch.json", extra_args=["--resume"]
+        )
+        self.assertEqual(argv[-1], "--resume")
+
+    def test_unsupported_client_refused(self):
+        with self.assertRaises(ValueError):
+            launcher.build_client_launch_argv("not_a_client", "/tmp/x.json")
+
+    def test_prepare_launch_writes_private_per_launch_config(self):
+        prepared = launcher.prepare_application_launch(
+            PROJECT_PROFILES,
+            client_type="claude_code",
+            namespaces=PROJECT_NAMESPACES,
+            launch_nonce="prep-1",
+            now=NOW,
+        )
+        path = prepared["launch_config_path"]
+        self.addCleanup(lambda: os.path.exists(path) and os.unlink(path))
+        self.assertTrue(os.path.exists(path))
+        self.assertEqual(os.stat(path).st_mode & 0o777, 0o600)
+        with open(path, encoding="utf-8") as handle:
+            payload = json.load(handle)
+        self.assertEqual(
+            sorted(payload["mcpServers"]),
+            ["gitea-author", "gitea-merger", "gitea-reviewer"],
+        )
+        self.assertTrue(prepared["shared_single_trusted_id"], prepared)
+        self.assertIn(path, prepared["argv"])
+
+    def test_two_prepared_launches_use_distinct_files_and_ids(self):
+        first = launcher.prepare_application_launch(
+            PROJECT_PROFILES,
+            client_type="claude_code",
+            namespaces=PROJECT_NAMESPACES,
+            now=NOW,
+        )
+        second = launcher.prepare_application_launch(
+            PROJECT_PROFILES,
+            client_type="claude_code",
+            namespaces=PROJECT_NAMESPACES,
+            now=NOW,
+        )
+        for prepared in (first, second):
+            path = prepared["launch_config_path"]
+            self.addCleanup(lambda p=path: os.path.exists(p) and os.unlink(p))
+        self.assertNotEqual(
+            first["launch_config_path"], second["launch_config_path"]
+        )
+        self.assertNotEqual(
+            first["client_instance_id"], second["client_instance_id"]
+        )
+
+    def test_unsupported_client_writes_no_config_file(self):
+        """Fail before creating a file, so a bad invocation leaves no residue."""
+        with self.assertRaises(ValueError):
+            launcher.prepare_application_launch(
+                PROJECT_PROFILES,
+                client_type="not_a_client",
+                namespaces=PROJECT_NAMESPACES,
+                now=NOW,
+            )
+
+
+class CliTests(unittest.TestCase):
+    """The runnable entry point itself."""
+
+    def _run_dry(self, argv):
+        import contextlib
+        import io
+
+        buffer = io.StringIO()
+        with contextlib.redirect_stdout(buffer):
+            code = launcher.main(argv)
+        return code, buffer.getvalue()
+
+    def test_dry_run_project_scoped_plan(self):
+        code, out = self._run_dry(
+            [
+                "--client",
+                "claude_code",
+                "--namespaces",
+                "author,reviewer,merger",
+                "--profile",
+                "author=prgs-author",
+                "--profile",
+                "reviewer=prgs-reviewer",
+                "--profile",
+                "merger=prgs-merger",
+                "--dry-run",
+            ]
+        )
+        self.assertEqual(code, 0)
+        plan = json.loads(out)
+        path = plan["launch_config_path"]
+        self.addCleanup(lambda: os.path.exists(path) and os.unlink(path))
+        self.assertEqual(plan["namespaces"], ["author", "reviewer", "merger"])
+        self.assertEqual(
+            sorted(plan["excluded_namespaces"]), ["controller", "reconciler"]
+        )
+        self.assertTrue(plan["shared_single_trusted_id"])
+        self.assertTrue(
+            fleet.assess_instance_identity(plan["client_instance_id"])["trusted"]
+        )
+        self.assertEqual(plan["argv"][0], "claude")
+
+    def test_cli_refuses_profile_for_unlaunched_namespace(self):
+        with self.assertRaises(SystemExit):
+            self._run_dry(
+                [
+                    "--namespaces",
+                    "author,reviewer,merger",
+                    "--profile",
+                    "author=prgs-author",
+                    "--profile",
+                    "reviewer=prgs-reviewer",
+                    "--profile",
+                    "merger=prgs-merger",
+                    "--profile",
+                    "controller=prgs-controller",
+                    "--dry-run",
+                ]
+            )
+
+    def test_cli_refuses_unsanctioned_namespace(self):
+        with self.assertRaises(SystemExit):
+            self._run_dry(
+                [
+                    "--namespaces",
+                    "author,admin",
+                    "--profile",
+                    "author=prgs-author",
+                    "--dry-run",
+                ]
+            )
+
+    def test_cli_refuses_missing_profile(self):
+        with self.assertRaises(SystemExit):
+            self._run_dry(
+                [
+                    "--namespaces",
+                    "author,reviewer",
+                    "--profile",
+                    "author=prgs-author",
+                    "--dry-run",
+                ]
+            )
+
+    def test_cli_refuses_malformed_profile_pair(self):
+        with self.assertRaises(SystemExit):
+            self._run_dry(
+                ["--namespaces", "author", "--profile", "prgs-author", "--dry-run"]
+            )
+
+    def test_two_cli_runs_receive_distinct_ids(self):
+        args = [
+            "--namespaces",
+            "author,reviewer,merger",
+            "--profile",
+            "author=prgs-author",
+            "--profile",
+            "reviewer=prgs-reviewer",
+            "--profile",
+            "merger=prgs-merger",
+            "--dry-run",
+        ]
+        _, first = self._run_dry(args)
+        _, second = self._run_dry(args)
+        a = json.loads(first)
+        b = json.loads(second)
+        for plan in (a, b):
+            path = plan["launch_config_path"]
+            self.addCleanup(lambda p=path: os.path.exists(p) and os.unlink(p))
+        self.assertNotEqual(a["client_instance_id"], b["client_instance_id"])
+
+
+if __name__ == "__main__":  # pragma: no cover
+    unittest.main()

tests/test_root_checkout_guard.py

@@ -87,13 +87,27 @@ class TestAssessRootCheckoutGuard(unittest.TestCase):
         self.assertTrue(result["block"])
         self.assertIn("tracked local edits", result["reasons"][0])
 
-    def test_head_behind_prgs_master_blocked(self):
+    def test_head_behind_tracking_base_ref_blocked(self):
+        """#983: the base ref is derived, so the message no longer hardcodes PRGS."""
         result = self._assess(
             head_sha=OTHER_SHA,
             remote_master_sha=MASTER_SHA,
         )
         self.assertTrue(result["block"])
-        self.assertIn("does not match prgs/master", result["reasons"][0])
+        self.assertIn(
+            "does not match the tracking integration ref", result["reasons"][0]
+        )
+
+    def test_head_behind_named_base_ref_reports_that_ref(self):
+        """The resolved ref is named, so a non-PRGS target is reported accurately."""
+        result = self._assess(
+            head_sha=OTHER_SHA,
+            remote_master_sha=MASTER_SHA,
+            remote_master_ref="refs/remotes/MDCPS/dev",
+        )
+        self.assertTrue(result["block"])
+        self.assertIn("does not match refs/remotes/MDCPS/dev", result["reasons"][0])
+        self.assertNotIn("prgs/master", result["reasons"][0])
 
     def test_merger_requires_clean_control_checkout(self):
         result = self._assess(