Review R1/N1 (reviewer-grok-01, planner-reviewer-codex-01), reproduced before
fixing. The usage line and SKILL.md both promise "relative to the control
root", but parse_args did Path(...).expanduser().resolve(), which is relative
to the caller's cwd. The runner never chdirs to workspace_root, so the two are
routinely different - the documented recipe runs the loop from the control root
while the source root is a worktree.
Two ways it went wrong: a plan that exists only under the control root was not
found at all (exit 1), and a same-named file in the caller's cwd was loaded
instead, quietly putting an unrelated plan into the first implementation
prompt. That second one is the very failure #25 is about, reintroduced through
the option meant to replace it.
Resolve against workspace_root the way jobs_dir already does (Issue #21 R2);
pathlib's `/` leaves an absolute or expanded-home path untouched. Every earlier
test passed an absolute tmp_path, which is why none of them caught this: P3c
covers both failure modes with the control root away from the caller's cwd, and
P3d pins that absolute and ~ paths are not rewritten.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The --plan-file check landed on the one path a human is watching. The three
--plan-mode loads read a report written by another agent with nobody watching,
and had no content check at all: find_report only tests that the path exists.
An empty or whitespace-only report therefore blanked current_plan in silence.
Phase 2's `if self.current_plan` then flipped the prompt to self-planning and
the run finished rc 0, having paid for a planner job, a Creator challenge and a
refine job and thrown the result away. The escalation refine was worse: it has
no guard, so the correction prompt asserted "follow the updated plan below" and
carried nothing.
All four loads share _plan_text() now - a plan is a readable regular file with
text in it - and each caller says what None means. An empty initial plan stops
the run instead of demoting it, and an empty refine warns and keeps the plan it
already has, which is still good. Those two together leave the escalation
prompt unable to claim a plan it does not carry, so it needs no new branch.
Found by the Creator's challenge to the plan (job 0f63ac1d), sustained by the
Planner as revision 1.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Without --plan, Phase 1 resolved a planner session automatically and globbed
<workspace_root>/.agents/reports/<planner>/plan-*.md for the newest file by
mtime, handing it to the Creator as "the plan already agreed on". Both keys are
wrong: the report directory belongs to a session, not to this task, so other
issues' plans pile up in it, and right after a clone or checkout mtime is
checkout order. 47 such plans are committed in this repository, so an ordinary
run could silently implement against an unrelated design.
Reuse is explicit now. --plan-file <path> is the only way to load an existing
plan; it is resolved against the control root, checked before a single job
exists, and refused if the file is missing, not a regular file, unreadable or
empty (an empty file would quietly turn into a "no plan" prompt). It cannot be
combined with --plan, and --review-only warns and ignores it because no
implementation job is delegated there. The implementation prompt itself is
unchanged, so the Creator's contract does not move.
Phase 1 now says which plan it loaded and where from, or that it loaded none.
The legacy shell loop (MAM_LOOP_PY=0) loses the same auto-load and rejects
--plan-file rather than duplicating the check on a path that is on its way out.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
- worktree_prepare.sh installs the same files deploy/install.sh does
(.agents/skills/multi-agent-mux-delegate-job/requirements.txt: paho-mqtt,
PyYAML) plus requirements*.txt. Without paho the loop wrapper, stop_session
and the subscriber tests fail in the worktree - seen as 4 suite failures
in a worktree without its own venv, all passing with one.
- review D1: a failed pip install removes the venv it half made, so the next
run recreates it instead of reporting VENV=kept (test w1c).
- review D3: the [worktree context] block's uncommitted-changes flag uses
--untracked-files=all like the guards.
- review D2: the create skill doc says the install needs the package index.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
R1/R2: `git status --porcelain` honours status.showUntrackedFiles, so a
`no` setting in a submodule (or the worktree) hid untracked files from the
checks that justify worktree_remove.sh's single --force and from the review
snapshot guard (_dirty_submodules), which then approved content it never
captured. All three checks (lib_py/worktree.py, worktree_remove.sh,
runner._dirty_submodules) now pass --untracked-files=all. The source tree's
own snapshot was already safe: `git add -A` ignores that setting.
R3: worktree_prepare.sh no longer links .venv to the control root's. It
creates the worktree's own venv with the control root's interpreter and
installs requirements*.txt; a leftover link is refused (exit 1), a control
root without an interpreter is a warning (VENV=skipped); VENV= is printed.
Tests: w1 (real venv, kept on re-run), w1b (link refused / skipped), w6e
parametrised over the hidden setting for the post-stop recheck, w6e2
(hidden untracked file in the submodule and in the worktree blocks the
removal), l17 untracked-hidden / during-hidden (guard refuses before the
first review; a mid-round hidden file makes the round stale). Each new
case fails without the fix.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
The 33 failures on the Linux / Python 3.14 host (#28) were environment
dependencies, not regressions; each test is worth keeping, so fix the
dependency instead of deleting the test:
- test_workspace_scope: `source` under `shell=True` ran in dash; run the
command string under bash.
- conftest autouse: MAM_ENV_FILE points at an absent file for every test,
so a workspace .mam.env with a reachable broker no longer breaks the
"broker unavailable" tests (T1-4).
- help-capture guard: compare normalized help text (whitespace collapsed,
"optional arguments"/"options" unified) so argparse re-wrapping across
Python versions is not drift.
- deploy freshness D-11..D-19 and D-22..D-30: skip with a reason when the
nats-docker submodule is not initialized.
- lib_py.agents.base.conversation_in_use and heartbeat._process_table:
`ps -eo pid=,ppid=,args=` (POSIX) instead of the BSD `-axo ... command=`
form, the most likely cause of the 12 purge failures on Linux (the holder
scan returned nothing); to be confirmed on that host.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
- review jobs' [워크트리 컨텍스트] points at the [리뷰 identity] path only (B-2);
remote re-adjudication gets a matching sentence (N-4)
- dirty/untracked submodule content fails the snapshot and makes a round stale,
regardless of submodule ignore settings (B-R1)
- with --worktree/--base and no reviewer, the creator self-review takes the
reviewer path: snapshot diff, evidence check, pre-promotion re-check and the
identity sidecar; NOT PASS still re-reviews without a correction job (B-R2/B-R3)
- snapshot diff failures keep the UNAVAILABLE marker (N-1); prune only when
worktree remove fails (N-3); the empty per-loop review dir is removed (N-5)
- tests L17-L21, B-2 assertions in L6/L11; SKILL.md env stale note; import order
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017R3Eg7MEpSWnm1oB9ZPP6G
Active with --worktree/--base only; without them prompts, flow and promotion
are unchanged (a loop_id is recorded in the checkpoint either way).
- Every review round is pinned (B3): the source's HEAD and its working state
as a tree built through a temporary index (the real index/status/refs are
untouched); a dirty state becomes a ref-less snapshot commit on HEAD
(committed as mam-loop, not as the user). An untracked nested repository
cannot be pinned and is refused. identity = (loop_id, base, head, tree).
- Review isolation (B4): a detached review worktree of the snapshot under
TMPDIR (default), or `--review-isolation env` (the source root with env and
temp-dir instructions). Failing to open the tree refuses the round - the
isolation is never lowered silently. Trees are removed after the round and
on any exit (run() wraps the body; release_guard too).
- The reviewers' diff comes from the snapshot (B13):
diff_collect.sh takes an optional head commit for a commit-to-commit diff;
its two-argument form is byte-identical.
- Stale rounds (B7/B8): after the round the source is measured again; if it
changed, the round's approvals are discarded and the next iteration
re-reviews without a correction job (last iteration: rc 1). Only the last
non-stale round's PASSes are promoted.
- Records (B9/B15): review_rounds in the checkpoint with per-reviewer
transport/evidence, and report-<jid>.identity.json next to each promoted
report: evidence review-tree | source-tree | diff-only.
- Pre-promotion re-check (B10): a source changed after approval is not
promoted (rc 1).
- Remote reviewers (B14, Creator challenge 91c81095): they cannot reach this
host's review tree. Checked before anything is delegated: refused by
default; `--remote-review diff-only` admits them with the snapshot SHAs and
diff as their only evidence (no local path in their prompt), and refuses a
round whose diff is truncated or unavailable.
- run_loop.sh accepts both options (usage identical to the runner's); the
legacy shell loop refuses them before the lock.
Tests tests/test_issue21_loop_b.py L1-L16 (22).
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017R3Eg7MEpSWnm1oB9ZPP6G
Both unpushed checks in lib_py/worktree.py - `submodule foreach` over the
initialised checkouts and the direct `--git-dir` walk over the repositories
under the worktree's modules/ - used `HEAD --branches --not --remotes`. A
commit made on a submodule's detached HEAD and kept only by a tag is reachable
from neither, the top-level status stays clean, and worktree_remove.sh then
deleted the repository with its single --force (exit 0).
Both checks now share one rev list, UNPUSHED_REVS = HEAD --exclude=refs/stash
--all --not --remotes: HEAD and every local ref (branches, tags, notes, ...)
minus remote-tracking refs. refs/stash stays out of it because the stash check
(stash list / reflog) owns it, so a stash is reported once, as 'stashes'. As
intended, a local-only note is now protected too. The stash, error, partial
repository and dirty checks are unchanged.
Tests W6l (tag-only commit; initialised submodule and deinit'd repository) and
W6m (annotated tag). The helper header, the stop SKILL.md and the design doc
state the wider rule.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017R3Eg7MEpSWnm1oB9ZPP6G
- N1 (planner-reviewer-claude-01) worktree_remove.sh: when `git status` fails
in the worktree (e.g. its directory was deleted), judge_tree returned before
METHOD was set and the summary line died under set -u with exit 1. METHOD
now has its default before the function; the refusal is exit 3 with
"nothing was changed", plus a `git worktree prune` hint when the directory
is gone. Same for --dry-run.
- R4 (planner-reviewer-codex-01) worktree_prepare.sh: a submodule path that is
not initialised yet but already holds files made the clone fail, and the
rollback's `deinit -f` then deleted those files. --submodules now refuses
such a path up front (new exit 3, nothing changed), which also makes the
rollback's premise true: after a failed attempt, everything under a
not-yet-initialised path was created by that attempt.
- R5 (planner-reviewer-codex-01) lib_py/worktree.py: `submodule foreach`
visits initialised checkouts only, but a deinit'd submodule keeps its
repository - stash and unpushed commits included - under the worktree's
modules/, which the forced removal deletes. Every repository under modules/
(nested ones included) is now checked directly (`--git-dir`): commits on no
remote (HEAD included) and refs/stash with its reflog. A directory that looks
like a partial repository cannot be checked and refuses.
- Also: the rollback loops are C-style (`seq 0 -1` counts down on BSD), and
prepare warns that --base is ignored for an existing branch.
Tests W6i (deleted worktree directory, with and without --dry-run), W6j and
W6j2 (stash / unpushed commit in a deinit'd submodule's repository) and W6k
(existing files in an uninitialised submodule path). Skill docs and the design
doc describe the new refusal and the wider repository check.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017R3Eg7MEpSWnm1oB9ZPP6G
Review ffb5055e (planner-reviewer-codex-01):
- R1 [P1] worktree_remove.sh: the clean-tree and submodule checks ran only
before --stop, so work a stopping session wrote (or committed, or stashed)
was deleted by the forced removal. The checks are one function now, run
before any change and again right before the removal; any finding then
refuses with exit 3 (sessions already stopped, worktree kept), and the
removal method follows that latest look.
- R2 [P1] lib_py/worktree.py: a submodule stash is reachable from no branch
and leaves the status clean, and the forced removal deletes the worktree's
own submodule repositories. `git stash list` in every submodule is a
separate finding (`stashes`), refused like dirty/unpushed; errors fail
closed as before.
- R3 [P2] worktree_prepare.sh: when one submodule cloned and a later one
failed, the rollback deleted modules/ while the cloned checkout's .git file
still pointed into it, and `git status` failed in the worktree. The rollback
now undoes only this attempt: each submodule whose git dir did not exist
before is deinitialised before its git dir is deleted; submodules
initialised earlier are kept; shared submodule.* config values are restored
exactly (new keys removed, pre-existing values put back - init and deinit
both write .git/config). A final `git status` either confirms a usable
checkout or prints what to inspect.
Tests W6e (a write during stop keeps the worktree and the file), W6f (a
submodule stash refuses), W6g (partial failure on a fresh worktree: usable,
no modules/, no new config, then prepare and a plain removal work) and W6h (a
failing retry keeps the earlier `sm` init and its config value). The skill
docs and the design doc describe the recheck, the stash rule and the scoped
rollback.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017R3Eg7MEpSWnm1oB9ZPP6G
- worktree_prepare.sh (create skill): idempotent `git worktree add` next to the
control root (a path inside it is refused), .venv linked to the control
root's, /.venv and .pytest_cache/ in the shared info/exclude, submodules only
with --submodules - a failed init removes its own traces (modules/, new
submodule.* keys in the shared config) and exits 1, since git refuses a plain
`worktree remove` once modules/ exists - and an optional creator registered
in the control root's .mam with MAM_NO_SUBSCRIBER_DAEMON=1. Prints BASE= for
`run_loop.sh --worktree ... --base ... --review-only`.
- worktree_remove.sh (stop skill): every check first, nothing changed on a
refusal (exit 3): linked and not locked, running sessions only with --stop,
no open job of those sessions except the delegate jobs stop cancels, clean
tree including submodules, and inside submodules no modification and no
commit on no remote (HEAD included - submodules sit on a detached HEAD).
Then stop, re-check jobs, and remove: plain, or a single --force when the
worktree holds submodule state.
- lib_py/worktree.py: the parsing behind both (stdlib only).
- A8: create's immediate reconcile and watchdog use the control root's code
and WORKSPACE_ROOT, not the worktree --workspace's copy.
Tests W1-W6d (local submodules allowed through GIT_CONFIG_* env only).
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017R3Eg7MEpSWnm1oB9ZPP6G
Issue #21 phase-1 prerequisites (PR #22) add loop runner options and the
control-plane module: MINOR. VERSIONS.md gains the v4.6.0 entry, the
archived design reviews lose their markdown trailing whitespace so
`git diff --check` is clean, and the three merge-side PASS reports
(claude, agy, codex @ 4a81d15) are archived under .agents/reports/.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
Review 9a1adb46 (planner-reviewer-codex-01):
- R1 [P1] lib.sh new-session: a forwarded binding reached the agent as
arguments. With env flags the idle-pane branch prefixed `env K=V ...` to
the command that `herdr agent start ... --kind <agent> -- <args>` passes to
the agent executable. An idle pane is now reused only when no environment
has to be set; otherwise the pane is created by pane split / workspace
create, which take --env. The dead `env` prefix code is removed (its
send-text fallback would have typed the same line).
- R2 [P2] runner: a relative MAM_JOBS_DIR is resolved once against the
control root, the way the wrapper (cwd=workspace_root) reads it, and that
absolute dir is used for delegation, wait, report lookup and cleanup.
- R3 [P2] stop: the row's control_plane.delegate_job_logs_dir is restored
(exported unless the caller set DELEGATE_JOB_LOGS_DIR), so the
cancellation history lands in the row's audit-log root.
Tests: T3-1 and T3-3 now assert that the binding arrives as --env on the
pane-creation calls and that no env/K=V token follows `--` in agent start
(fresh and reused pane); T3-4c (audit root on stop) and T2-15 (relative
MAM_JOBS_DIR) are new. The design doc's compatibility table and a create
comment no longer describe the removed idle-pane env path.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017R3Eg7MEpSWnm1oB9ZPP6G
A session row now records the control plane it belongs to, and the pane
gets it explicitly instead of relying on env inheritance.
- lib_py/control_plane.py (+ thin lib.sh wrappers mam_control_plane_json /
mam_control_plane_prefix): the binding is WORKSPACE_ROOT,
AGENT_SESSIONS_YAML, MAM_JOBS_DIR, DELEGATE_JOB_LOGS_DIR as absolute paths.
- create: the binding is always persisted on the row (`control_plane`) and
forwarded to the pane only when it differs from what --workspace alone
implies, as leading K=V tokens that lib.sh's new-session already turns into
--env (fresh pane) or `env K=V` (reused idle pane). pane.cmd_full and
start_command stay unprefixed. A binding with whitespace that would have to
be forwarded is refused with exit 2 (lib.sh splits the flags unquoted); the
~/.local/bin wrapper path records the binding but warns it is not forwarded.
- resume: the row's binding wins over the caller's env (a difference is
reported on stderr) and is forwarded again; a legacy row without one uses
the caller's binding and update_yaml_resumed.sh records it.
- stop: registry order MAM_JOBS_DIR > the row's mam_jobs_dir >
$WORKSPACE_ROOT/.mam/jobs > .mam/jobs, exported for delegate_publish_event.
- status: both get_job_status copies look in MAM_JOBS_DIR, the row's
registry and the matching audit logs before the old candidates.
- reconcile: MAM_JOBS_DIR wins in the subscribe heredoc and is passed in.
Tests T3-0..T3-8 (mock herdr: forwarding is verified at the herdr-call level
only; real pane env propagation is left to item 7). T3-7 is a static guard.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017R3Eg7MEpSWnm1oB9ZPP6G
- --worktree <path> (CLI over MAM_LOOP_WORKTREE): the source root. Only the
diff, base, head and the empty-set check use it; the wrapper, registry,
checkpoint, loop guard, promotion and cleanup stay on the control root. The
collector is the control root's diff_collect.sh (never the worktree's
unreviewed copy), its arguments shlex-quoted. Validation: an existing
directory, the top of its working tree, the same git common dir as the
control root (".git" is resolved against the directory it came from).
- --base <ref> (CLI only, no env - a stale sha would silently widen the next
range): resolved to a full sha, must be an ancestor of the source HEAD.
- --review-only: requires --base, refuses --plan; Phase 2 is skipped and
logged; correction and comply jobs still go to the creator.
- With --worktree or --base every delegated prompt starts with a context
block (path, branch, base and its origin, head, base..head commit count,
uncommitted changes); without them prompts are byte-identical. All call
sites in run() go through one _dispatch helper.
- Fail-closed guard (only with --worktree/--base): before each review round,
self-review included, an empty change set or one git cannot determine
(exit codes, never output parsing) ends the loop with rc 1 instead of
asking for reviews. This closes the fail-open where a loop started after
an out-of-loop commit sent reviewers "(no changes since base commit)".
- The runner honours MAM_JOBS_DIR and passes --registry-dir to the wrapper.
- run_loop.sh accepts the three options; the legacy shell loop
(MAM_LOOP_PY=0) refuses them before the freeze and the loop lock.
Tests T2-1..T2-14.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017R3Eg7MEpSWnm1oB9ZPP6G
Review round 5 (planner-reviewer-codex-01, NOT PASS at 2b1d590):
- F1 bounded lock cycle: the stopper holds the per-job lifecycle lock while
it waits for the timer to exit, and the timer's exit cleanup waited for
that same lock, so every stop ran out the drain and ended in SIGKILL with
a false "did not drain" warning. The exiting timer now tries the lock
without blocking (lifecycle_lock(blocking=False)); when it is busy the
ownership operation holding it removes the pidfile, and when it is free
the own-pid comparison is kept.
- F2 untracked owner: when writing the new timer's pid fails after the
spawn, _spawn_timer_locked kills and reaps that exact process while the
lock is still held and drops only a partial pidfile of that attempt,
then re-raises.
- tests: an idle timer stops well before a 10 s drain without SIGKILL;
a pidfile failure after spawn leaves no survivor, no pidfile and a free
lock (both fail on 2b1d590); a spawn failure leaves nothing behind and
the next start works.
- docs: MESSAGING 2.5 and VERSIONS item 1 describe the lifecycle lock;
VERSIONS test count 31.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGpJXhSZasJLU38ksDghe
Review round 4 (codex R3): two concurrent `start_timer` calls for the same
attempt both passed the "no timer" check, both spawned, and the pidfile
tracked only one - the untracked timer could still publish a retained
`running` after `completed`. A per-job flock (`<job>.heartbeat.lock`,
separate from the registry/seq lock so in-flight beats can still take
their seq) now makes observe/drain/spawn/publish-pid one ownership
operation; the terminal stop takes the same lock; a start that finds the
attempt ended (terminal record or raised barrier) does not arm; the timer
removes its own pidfile only under the lock so a replacement's pid is never
deleted. Regressions: concurrent starts leave exactly one tracked owner that
the terminal stop drains; a start after the barrier arms nothing.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
Review round 3 (planner-reviewer-codex-01, NOT PASS R3 at 7e1d3f0):
start_timer stopped a running timer with a bare TERM and overwrote the only
pidfile. A beat the old timer had already admitted - seq taken, recorded on
disk, held at the wire - stayed alive but undiscoverable, so a later
terminal publish drained only the replacement and the old beat could still
put a retained `running` on the broker after `completed`.
- heartbeat.start_timer: stop_timer(drain=MAM_HEARTBEAT_DRAIN_S). The old
timer is drained or killed, and confirmed gone, before the pidfile names
the new one. A terminal publish racing the replacement either drains the
same old pid or meets a new timer whose beats the barrier refuses.
- test: an old timer held at the wire after admission is replaced, then
the job completes; the old timer is gone before the handover and nothing
it held follows the terminal (fails on 7e1d3f0).
- docs: MESSAGING 4.3 no longer says any state resets the idle clock;
MESSAGING 2.5 and VERSIONS state the replacement rule; test count 26.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGpJXhSZasJLU38ksDghe
Review round 2 (planner-reviewer-codex-01, NOT PASS at d09c162):
- R1 remote.mirror_job: admission under the registry lock. The same
dispatch_token is idempotent (a finished attempt never runs again), a
lower attempt or the same attempt under another token is refused before
any write, and only a strictly newer attempt replaces the record
(register_job(supersede_older_attempt=True); events.log kept, seqs stay
monotonic).
- R2 remote.handle_dispatch: the --exec command runs in its own session;
at timeout_sec, or when the runner is interrupted (SIGTERM now unwinds
it), the whole process group is terminated (TERM, then KILL) before the
error is published.
- R3 terminal barrier: a terminal publish raises an attempt-scoped barrier
under the lock that allocates seqs, stops the heartbeat timer, gives it
MAM_HEARTBEAT_DRAIN_S (default 1 s) to exit and then kills it, and only
then takes its own seq. Beats are admitted under that lock
(next_seq(admit=...)); a beat still on the wire is cancelled with the
timer rather than holding the terminal back; and a late non-terminal
event never publishes a running state over a terminal record.
- R4 job_subscriber: a retained state renews the idle clock only when its
seq is newer than everything already accepted; re-delivered or older
snapshots still recover a terminal status.
- tests: five regressions, all failing on 77f2829; T1-14's wall clock
15 -> 30 s against public-broker latency.
- docs: MESSAGING 2.5 (barrier, state renewal, detection vs final close),
heartbeat docstring, loop SKILL.md (remote rows need the Python runner),
VERSIONS v4.5.0 items 1/4/5 and the test count (25).
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGpJXhSZasJLU38ksDghe
In the v4.5.0 review round grok read the brief and published `started`
70 s after the injection; with T = 60 the wrapper had already closed the
job as agent_blocked and refused its later PASS. T bounds delivery, not
work, so a generous default costs nothing when the agent answers fast.
Code, docs and the CLI snapshot updated together.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
Review round 1 (claude NOT PASS) corrections:
- B2: the timer now watches a process - `--watch-pid`, or by default the
nearest agent-CLI ancestor of the publishing command - and stops within
one period once it is gone, so a crashed agent stops heartbeating and is
closed after K x N; a remote runner watches itself. Documented as process
liveness (an alive-but-stuck agent keeps heartbeating).
- R1: `_wait_started` accepts an events.log `started` only when it carries
the current dispatch token (loop jobs share one events.log).
- R2: the remote runner ignores a re-sent dispatch of an attempt that
already reached its terminal.
- R3: registry.md EOF blank line (git diff --check clean).
- R4: the ack error states the effective budget when the wall clock cut it.
- P3: stop_timer verifies the pidfile pid is this job's timer before
signalling; a retained state message keeps its own seq watermark so it
can never make the monotonic guard drop its own event.
- MULTI_AGENT_RULES.md timeout section updated for the contract.
- tests: agent-process death stops the timer, recycled-pid safety,
attempt-scoped started, re-sent dispatch; acceptance 3 now kills the
agent process instead of the timer. Help captures refreshed
(--watch-pid). The wrapper can be sourced with
MAM_DELEGATE_JOB_NO_MAIN=1 for helper-level tests.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
Contract (no new event types): heartbeat every N s (automatic timer),
unresponsive after K x N s without an event, `started` within T s of the
injection, retained per-job state, and a remote transport.
- registry: heartbeat_sec / heartbeat_misses / ack_timeout_sec / transport
fields; idle_timeout_sec derives K x N unless explicit (defaults 30/4/60
keep the historical 120 s idle window).
- heartbeat.py + publish_event --heartbeat N: detached fixed-cadence timer
publishing signed `progress` heartbeats (threaded wire publish, local
record first); stopped by the terminal event via pidfile, by re-dispatch
and by lifetime. Terminal events refuse --heartbeat.
- publish_event: retained `<prefix>/state` (status, last_event, seq, HMAC)
sent on the same connection after every event (--no-state to skip);
publisher.publish_raw takes extra messages.
- job_subscriber: subscribes the state topic, recovers terminal status from
retained state, resets idle on any accepted message, records seq_gap and
reconnected in the audit log, idle budget before the first event is
T + K x N (--ack-timeout), --remote mirrors started/terminal events into
the registry through the transition boundary.
- lib.sh send_keys_safe: an accepted-but-unconfirmed submission is confirmed
by agent_status working or the job's own `started` (_sks_job_started);
pane change and busy token are no longer delivery signals.
- wrapper: --heartbeat/--heartbeat-misses/--ack-timeout (MAM_* env),
injected start command arms the timer, brief states the contract,
_wait_started (registry / events.log / subscriber log, bounded by T and
the wall clock) replaces the 10 s inject grace, R1 grace uses only the
on-disk event age (herdr/pane logged once as a diagnostic),
remote:<name> sessions skip preflight/idle/pane and dispatch over the
broker (remote_dispatch).
- remote.py (+ scripts shim): signed retained dispatch (MAM_DISPATCH_SECRET)
and a runner that mirrors the job into its own registry with the
orchestrator's identity, publishes started --heartbeat, runs --exec and
publishes the terminal event.
- loop runner: passes MAM_HEARTBEAT_SEC / MAM_HEARTBEAT_MISSES /
MAM_ACK_TIMEOUT_S through; YAML rows with transport: remote delegate as
remote:<name>.
- docs: MESSAGING.md 2.5 contract + remote transport, delegate-job SKILL.md,
registry.md, loop SKILL.md.
- tests: tests/test_issue18_liveness_contract.py (16); issue 14 pane-signal
tests flipped to started-based; issue 6 working-agent tests acknowledge
with started; CLI snapshot/help captures refreshed.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
create_session.sh detaches `reconcile.sh --subscribe --idle-timeout 0` per
workspace; test sandboxes never reaped it and the daemon neither noticed its
workspace vanish nor honoured SIGTERM (bash deferred the signal behind the
foreground python child). 202 orphans were found holding public-broker
wildcard subscriptions for deleted pytest tmpdirs.
- daemon.py / lib.sh: MAM_NO_SUBSCRIBER_DAEMON=1 (literal) skips the spawn but
still reports a running daemon; cli prints a stderr note, stdout stays empty.
- reconcile.sh: python subscriber runs as a waited background child with a
TERM/INT trap (exit 143); the MQTT loop exits when the workspace dir is gone
or .mam/monitor.lock is no longer the held inode; SIGTERM/SIGINT handled via
the stop event so loop_stop/disconnect run; the polling fallback checks the
same and sleeps in a waited child.
- conftest: sandbox default MAM_NO_SUBSCRIBER_DAEMON=1, per-sandbox reaper
finalizer, session-scope leak guard over the pytest basetemp.
- tests: 11 lifecycle tests (tests/test_issue19_subscriber_daemon_lifecycle.py);
phase-1 watchdog/daemon characterization tests opt back in.
- docs: monitor SKILL.md lifecycle contract + pitfalls, create SKILL.md note.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
Review of 4176d75 (planner-reviewer-claude-01, N1): _cmd_scrub_secrets chose
records by the presence of broker.password, not by its value. The old
to_registry_block wrote "password": null even for an unauthenticated broker,
so on this repo's registry (901 of 903 records carry the null key, none a
real password) --dry-run reported "would scrub 901 record(s)" and a real run
would rewrite every one of them while holding the registry lock.
- registry.py: select on broker.get("password"); an empty key is no secret,
and _atomic_write_record drops it on the record's next write anyway
- tests: the migration test gains a legacy "password": null record that must
be neither listed nor rewritten (fails on 4176d75 with "would scrub 2")
- VERSIONS.md v4.4.4 item 2 and registry.md now describe the stored-value
rule
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGpJXhSZasJLU38ksDghe
PATCH release for the fix/issue-12-8-7 branch (Issue #12 broker password
kept out of job records, #8 fresh install stamps kept on update, #7
same-cwd workspaces packed). No code logic changes.
- lib.sh MAM_VERSION, lib_py/loop/eager.py FRAMEWORK_VERSION and the eight
SKILL.md frontmatters (create, stop, resume, status, monitor, delegate-job,
loop, orc-onboard) move from 4.4.3 to 4.4.4
- VERSIONS.md: current release v4.4.4 (2026-09-13 KST), the eight-row skill
matrix, and a v4.4.4 changelog section in the v4.4.3 format covering the
password-free registry block restored from MQTT_PASSWORD at connect time,
the scrub on every record write, `registry.py scrub-secrets [--dry-run]`,
why auth_token/dispatch_token stay, the update.sh stamp and install_state
merge rules, the sibling-workspace walk in new-session, and the 8 new tests
- VERSIONS.md v4.4.3 ledger correction: its new-test count predated fae70c1,
which added two purge tests; now 26 (19 + 7) with both listed
- README, README.ko, BOOTSTRAP and BOOTSTRAP.ko badges and the MESSAGING.md
"Applies to" header follow the current version; historical 4.4.3 mentions
(v4.4.3 changelog, reviewer reports) stay as history
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGpJXhSZasJLU38ksDghe
Issue #12 - MQTT_PASSWORD was copied into every job record, which any
sub-agent can read. BrokerConfig.to_registry_block no longer persists it;
broker_config_from_job layers the password back from the environment
(.mam.env) at connect time, so publish_event still connects from the record.
_atomic_write_record scrubs broker.password from any record it writes, and
`registry.py scrub-secrets [--dry-run]` migrates an existing registry.
auth_token/dispatch_token stay: the agent's own publish_event needs them.
Issue #8 - update.sh restored the pre-update asset_hashes.txt and version.txt
over the installer's fresh ones, so version.txt named the previous commit and
remove.sh backed up every pristine updated file as "locally modified". The
success path now keeps the installer's stamps and only merges install_state
keys the re-install did not write (gitignore_created); the failure path
still restores everything verbatim.
Issue #7 - the shim's new-session took the first same-cwd pane, always found
the full first workspace, judged it overflow and opened another workspace per
agent (4 + 1 + 1). It now walks every same-cwd workspace in pane order and
takes the first with an idle pane or room in its 2xK grid; a workspace is
created only when all are full (4 + 2).
Tests: test_issue7_workspace_packing.py (3; the mock herdr gains an opt-in
per-workspace 2x2 geometry), test_issue8_update_state.py (1, real
install+update cycle), test_issue12_registry_secrets.py (4); registry CLI
snapshot and help capture refreshed for the intentional scrub-secrets
subcommand; registry.md documents the broker block.
Refs #12#8#7
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
Round-3 review of f5a6575 (planner-reviewer-codex-01, R1/R2).
R1 - the adapter's in-use check handed only the conversation DB to lsof, so a
holder that keeps the presence lock (or a sidecar) open while naming the uuid
on no command line and in no session variable was invisible, and the held
lock was unlinked from under it. AgyAgentAdapter.conversation_in_use now
passes every path of the conversation (DB, -wal/-shm/-journal, brain,
annotation, presence lock).
R2 - both resolver stages required the DB, so once an interrupted or pre-#10
purge had removed it, the WAL/annotation/lock left behind were "no artifact"
(exit 5) forever. BaseAgentAdapter.has_artifacts (default: a uuid-bearing
artifact path that exists) replaces the resolver's path fallback, and agy's
override accepts any of its pieces.
Tests: a flock holder that names the uuid nowhere is refused and nothing is
removed; orphaned leftovers resolve to agy and are purged. VERSIONS.md
v4.4.3 items 1-2 and orc-onboard SKILL.md updated.
Refs #10#13
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
Review of 0282312 (planner-reviewer-claude-01, N1): conversation_in_use
excluded the purge's whole ancestor lineage, and that lineage includes the
agent CLI that launched the skill. An orchestrator purging its own UUID
from its own session was therefore never refused - the only process naming
the UUID was excluded, and claude keeps no transcript open for lsof - so
its live conversation was deleted, contrary to what orc-onboard SKILL.md
promises.
- base.conversation_in_use: in use when one of the caller's session-id
variables (CLAUDE_CODE_SESSION_ID, ANTIGRAVITY_CONVERSATION_ID,
HERMES_SESSION_ID, GROK_SESSION_ID, OPENCODE_SESSION_ID) equals the uuid;
the command-line check now excludes only the purge's own helper chain and
stops at the nearest agent CLI ancestor (basename claude/agy/hermes/grok/
opencode/codex, read like detect_nearest_agent), which counts as a holder
when it names the uuid; the lsof check is unchanged
- tests: the non-agent helper chain is excluded but an agent ancestor is
counted; the purge CLI is refused under an agent CLI naming the uuid and
with each of the five session variables set to it; orc_onboard --remove
--purge-conversation --yes is refused from the orchestrator's own session
(exit 5, registration kept). The eight new cases fail on 0282312
- job_subscriber --registry-dir now defaults to default_registry_dir() like
the other entry points (review R-d)
- orc-onboard SKILL.md spells out how the calling session is recognised;
VERSIONS.md v4.4.3 item 2 and the new-test count (24)
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGpJXhSZasJLU38ksDghe
PATCH release for the fix/issue-10-11-13-15 branch (Issue #10 agy purge
sidecars, #11 orc_onboard message, #13 orchestrator conversation purge,
#15 workspace-anchored job registry). No code logic changes.
- lib.sh MAM_VERSION, lib_py/loop/eager.py FRAMEWORK_VERSION and the eight
SKILL.md frontmatters (create, stop, resume, status, monitor, delegate-job,
loop, orc-onboard) move from 4.4.2 to 4.4.3
- VERSIONS.md: current release v4.4.3 (2026-09-13 KST), the eight-row skill
matrix, and a v4.4.3 changelog section in the v4.4.2 format covering the
agy purge sidecars/annotation/presence lock and residual_artifacts,
`lib_py.agents purge` and `orc_onboard.sh --remove --purge-conversation`
with their exit codes, the past-tense message, the MAM_JOBS_DIR ->
WORKSPACE_ROOT -> cwd registry/log resolution and the 20 new tests
- README, README.ko, BOOTSTRAP and BOOTSTRAP.ko badges and the MESSAGING.md
"Applies to" header follow the current version; historical 4.4.2 mentions
(v4.4.2 changelog, reviewer reports) stay as history
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGpJXhSZasJLU38ksDghe
Issue #10 - agy purge_artifacts left resumable pieces behind: it now also
removes the SQLite sidecars (-wal/-shm/-journal), the title annotation and
the presence lock (the lock only once no live process references the
conversation), and residual_artifacts() reports anything under
~/.gemini/antigravity-cli still carrying the uuid.
Issue #13 - the orchestrator has no herdr_sessions row, so its conversation
could not be purged by any skill. `lib_py.agents purge <agent|auto> <uuid>`
resolves the adapter from what is on disk (orchestrator_uuids stores bare
uuids), refuses while a live process names the uuid (conversation_in_use:
process list minus the caller's own lineage, plus lsof), and prints what it
purged. `orc_onboard.sh --remove <uuid> --purge-conversation --yes [--agent]`
runs it before the registry update so a refusal changes nothing (exit 4
without --yes, 5 while in use).
Issue #11 - "Successfully removeed": conjugate add/remove explicitly.
Issue #15 - the job registry and audit log followed the caller's cwd, so the
pytest suite (create_session.sh with its default onboarding job, run from the
repo root) had left 593 test records in the real .mam/jobs. Defaults now
resolve MAM_JOBS_DIR / DELEGATE_JOB_LOGS_DIR, then $WORKSPACE_ROOT/.mam/...,
then the cwd (mqtt_common.default_registry_dir / get_logs_dir; registry.py,
publish_event.py, service.py, create_session.sh, stop_session.sh); the
sandbox fixture pins both. The tier suites that used to leak now leave the
real registry untouched (881 records before and after).
Tests: test_issue10_13_purge.py (13), test_issue15_registry_isolation.py (7);
T-O13 pins the corrected message; the R1 helper test and B-9 logs-dir test
now assert the workspace-anchored contract; publish_event snapshot updated.
Refs #10#11#13#15
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
Round-2 review of 952657e (reviewer-grok-01): R1's disk-heartbeat
evidence comes from events.log, but publish_event appended to it only
after with_retry(_publish_once) returned - after the broker ACK or all of
its retries. With the public broker at ~5 s per publish, an agent's
heartbeat reached the disk after the grace window and the live job was
closed as deadline_timeout (test_wrapper_idle_timeout_grace_accepts_disk_heartbeat
failed 2/2 on 952657e here, as it did for the reviewer).
- publish_event: new record_local_event appends the prepared payload to
events.log right after prepare_event (validation and seq allocation
first) and before the wire publish; record_outcome takes
local_event_recorded=True so the line is not written twice, while the
audit "published" entry and the status sync stay after the publish
- DelegationService.publish_event: the same local-first order
- wrapper comment and VERSIONS.md v4.4.2 item 6 now say the heartbeat is
on disk before the publish; new-test count 21
- tests: the heartbeat is already in events.log when the publish starts
(and is recorded once), and a publish failing on every retry leaves one
local record with published=False in the audit; both fail on 952657e.
The wrapper disk-heartbeat test keeps the real publish_event path and
now passes 3/3 at the same broker latency
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGpJXhSZasJLU38ksDghe
PATCH release for the fix/issue-9-14 branch (Issue #9 herdr server env
isolation, Issue #14 accepted-prompt delivery confirmation). No code logic
changes.
- lib.sh MAM_VERSION, lib_py/loop/eager.py FRAMEWORK_VERSION and the eight
SKILL.md frontmatters (create, stop, resume, status, monitor, delegate-job,
loop, orc-onboard) move from 4.4.1 to 4.4.2
- VERSIONS.md: current release v4.4.2 (2026-09-12 KST), the eight-row skill
matrix, and a v4.4.2 changelog section in the v4.4.1 format covering the
server env marker denylist, MAM_NO_AUTOSTART, the shim ls "no server
running" translation, SKS_DELIVERY_GRACE_S, MAM_INJECT_ACK_GRACE_S, the R1
pane busy-token grace, the T1-7 trigger change and the 13 new tests
- README, README.ko, BOOTSTRAP and BOOTSTRAP.ko badges and the MESSAGING.md
"Applies to" header follow the current version; historical 4.4.1 mentions
(v4.4.1 changelog, IMPROVEMENTS.md ledger, docs/contracts/issue-6, reviewer
reports) stay as history
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGpJXhSZasJLU38ksDghe
Issue #5 (Python core migration, CAS state transitions, shadow watcher,
auto-completion gate, eager-import immutable runtime, B-13 freeze removal)
is complete and merged to main. This stamps the release.
SemVer: MINOR, not MAJOR. The change is large but breaks nothing. Across the
whole Issue #5 range there are zero file deletions and zero renames, the event
wire schema stays at SCHEMA_VERSION 1, all four literal script paths survive as
compatibility shims, the 968-line legacy shell loop is still in run_loop.sh,
and every default change reverts through a single environment variable
(MAM_DELEGATION_PY / MAM_ENGINE_CAS / MAM_LOOP_PY / MAM_LOOP_FREEZE). The
riskiest new feature, auto-completion, ships default-off. No operator has
anything to migrate, which is the decisive reason not to call it 5.0.0. The
reasoning is recorded in the VERSIONS.md release notes, not just here.
Close a gap in the lockstep guard. lib.sh exports MAM_VERSION and eager_load()
refuses to start the loop when it disagrees with FRAMEWORK_VERSION, so eager.py
is a load-bearing fourth leg — but test_version_consistency.py only checked
three, and the sole check on eager.py was a hardcoded literal in the phase 4
test that a release bump has to edit anyway. The lockstep test now covers
eager.py, a new test pins the fail-closed gate itself, and the phase 4 test no
longer carries a version literal.
Fix a latent test-isolation defect the bump exposed. The first full run failed
four tests with "MAM_VERSION mismatch: expected 4.4.0, got 4.3.0 in
environment" — 4.3.0 was not in the tree, it was inherited from the shell that
launched pytest and had sourced the pre-bump lib.sh. Those tests were measuring
the runner's shell rather than the tree under test, and only passed until now
because the two values always happened to agree. A session-scoped fixture drops
the ambient value. Production is unaffected: run_loop.sh:17 sources lib.sh
before anything else and lib.sh re-exports MAM_VERSION unconditionally, so a
stale inherited value is always overwritten before Python starts.
Full suite: 750 passed in 912.50s, 0 failed, 0 skipped.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UrDNJkHmm18uu7n7JwV5Qd
The subscriber marked a notification intent as seen before the monotonic
sequence guard ran. A late packet was then dropped by that guard with its
intent already burned, so the genuine retry of a notification this
subscriber had never delivered was mistaken for a replay and dropped too.
Identity comparison stays ahead of the sequence guard -- a duplicate must
not consume a sequence number -- but the bookkeeping write moves behind
it, so "seen" means "accepted" rather than "arrived".
An intent that cannot yield a stable identity (not an object, or holding
an unhashable value) is now refused instead of raising out of the
callback: a paho callback that raises takes the network loop down for
every job that subscriber is watching.
Six regressions; three of them fail on the unfixed tree, and one drives
a real notifier retry through the real consumer with no hand-built
payload. Full suite: 732 passed, no skips.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UrDNJkHmm18uu7n7JwV5Qd
Two findings from round 2, both reproduced before being accepted.
R2 -- attribution was compared before the terminal write and nothing tied the
two together. A job could complete, be re-dispatched and claimed again in that
window, and evidence from the first attempt then satisfied a status check and
finished the second one. Reproduced with real transitions: committed, with the
registry already on iteration 2.
This corrects my own earlier reasoning, and I want to be plain about it. Last
round I argued that the attempt guard could live entirely in the evidence layer
because that is where attempt identity lives. That holds for comparing evidence
to a dispatch, and it does not hold for atomicity: a precondition can only be
atomic with the commit if the transaction can read it under its own lock. So the
identity now lives in the record.
It is an opaque dispatch token, not the iteration counter. That distinction is
the part of the earlier rebuttal that still stands: iteration already means the
loop round, and comparing a dispatch attempt against it produces a guard that is
vacuous when they agree and wrong when they diverge. A token has no competing
meaning. It is minted when an attempt begins -- claim and redispatch -- and the
gate passes the token it decided against, so any intervening re-dispatch refuses
the commit.
R4 -- the intent identity was being emitted but nothing consumed it. A replayed
notification is re-published with a fresh wire sequence, so the monotonic guard
cannot recognise it; to that guard a retry looks new. The subscriber now
deduplicates on the stable intent identity, leaving the sequence guard as the
rule for ordinary events. Across a restart the disk fallback resolves terminal
status and a duplicate terminal is refused as an idempotent ack, so a replay
after restart cannot double-apply either.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UrDNJkHmm18uu7n7JwV5Qd
The important correction is an honesty one, raised independently by two
reviewers. The note claimed startup recovery runs in the subscriber. It does
not: nothing in production imports the gate or the recovery function, and there
is no per-job opt-in path, so the flag is currently a no-op in the live loop.
Phase 2b delivers a verified library and gate; wiring it in belongs to phase 3,
and the note now says so and lists what wiring requires.
Also records the checks that are now mandatory rather than caller-dependent,
that the rollback switch disables auto-completion too, that backend keys match
exactly, that bookkeeping may touch one field, that settling is conditional on
the intent actually sent, and why the intent's identity travels in the payload
rather than relying on the transport sequence.
Fixes three stale references: task.md named the pre-move path for finalize, the
loop package description still described itself as shadow-only, and a comment in
mqtt_common still pointed at the engine's original location.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UrDNJkHmm18uu7n7JwV5Qd
Seven blocking items across three reviews, every one reproduced here before
being accepted. Two themes run through them: a gate whose strictness depended on
what the caller remembered to pass, and a boundary that trusted its only current
caller instead of enforcing its own rule.
Independent verification is no longer optional. Omitting the report directory
skipped the disk check entirely and handed the decision back to the collector's
own flags -- the self-certification this design exists to prevent. It is now a
refusal, as is an empty directory or a job with no recorded dispatch.
Evidence is now bound to its target. The policy compares a scenario against the
dispatch context carried inside that same scenario, which is self-consistent by
construction, so one job's evidence could complete another. The gate now checks
the job, attempt, session and pane generation against the recorded dispatch.
A reviewer with no verdict token is no longer completed. Phase 0 made this a
safety contract and run_loop reads a missing token as NOT PASS, so inferring
completion first would start a correction round against a review that was never
finished. The requirement applies to reviewers only.
Bookkeeping may write one field. A status-free request was an unrestricted merge
on a terminal record: it could rewrite the prompt and the session. Settling is
also conditional now on the exact intent that was sent, because publishing
happens outside the lock and a newer intent can legitimately land mid-flight;
clearing unconditionally dropped a notification nobody had sent. The intent's
own identity travels in the payload, since the transport sequence cannot
identify a retry of the same intent.
Side-effect evidence must be evidence. A kind with no reference names something
without evidencing it, and the CLI parser could not be the only check because
the API and hand-written manifests bypass it.
Also honours MAM_ENGINE_CAS=0, which the operations note already promised
disables auto-completion, matches backend keys exactly rather than by prefix so
a future name cannot inherit evidence never collected for it, records an
absolute report path, and imports the notifier at module scope.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UrDNJkHmm18uu7n7JwV5Qd
Two failures the full suite caught, both mine.
finalize.py started in lib_py/loop, which made publish_event import a sibling
package. That is the rule phase 1 wrote down and phase 2a already made me move
the engine for: under a frozen loop, delegation is bound to one tree while the
parent may be another, so reaching sideways mixes them. The manifest is written
into the job's own report directory and the CLI that declares it is a delegation
entry point, so finalize belongs there. Second time on the same rule, and the
guard caught it both times.
The publish_event --help golden no longer matched, because phase 2b
intentionally added the finalization options. The diff is exactly those three
options and nothing else, so the capture is refreshed rather than the assertion
weakened, and the test now states when a refresh is legitimate: an intended
extension, never drift nobody meant.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UrDNJkHmm18uu7n7JwV5Qd
Three scenarios join the contract set, so the fifteen vectors now cover an
unproven side effect, an undecidable attribution, and an agent whose state goes
unreadable in the window between finalization and notification. All fifteen
match their expected decision.
The side-effect case needed a decision about layering. Authoritative enforcement
belongs in manifest verification, which has the job record and therefore knows
what was declared; the policy honours an explicit negative from the collector so
the case is expressible as a vector. Absence is not read as proof there, because
the manifest check is the layer that fails closed on missing evidence.
The operations note records why the agent declares and the reader re-derives,
the three gates and why the backend allowlist can only be narrowed, the
notification transaction, and the rollback.
It also records a deviation from my own plan. §D2 asked for reviewer sign-off
before promoting an attempt id into the record, since that changes a contract
phase 1 froze. Measurement during implementation showed the change was not
needed: the attempt guard already lives in the evidence layer, where attempt
identity actually is, so auto-completion passes the engine a status precondition
instead. The schema is untouched, the frozen contract holds, and a
characterization test fails the moment a record starts carrying an attempt id,
which is when to revisit it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UrDNJkHmm18uu7n7JwV5Qd
The gate does not re-decide anything. Phase 2a already decides, fail-closed and
against contract-pinned vectors; this supplies real evidence to that decision and
commits only if three independent gates agree: the global switch is on, the
backend has observed evidence that it reaches a quiescent state, and the job
itself opted in. Default is off.
The backend allowlist is evidence-based rather than optimistic. Inferring
"finished" from stillness requires stillness to have been observed, and only
three of six backends have that. An operator may narrow the list; a name that
was never observed is dropped rather than trusted, so it cannot be widened by
configuration.
Everything else holds. An unreadable manifest, a report rewritten after
finalization, missing evidence for a declared side effect, a policy that raises:
all refuse. "Could not decide" is never "done". A late explicit terminal wins,
because the boundary refuses the inferred one -- that is the boundary working,
not an error in the gate.
Notification intent is committed in the same atomic write as the state, so a
crash between committing and publishing leaves a record that still says the
world has not been told. Recovery re-sends; duplicates are already dropped by
sequence. A failed send deliberately leaves the intent pending, because an
unsent notification that looks sent is worse than one that is retried.
Two fixes this surfaced. Settling used a request that terminal records refuse,
which made an owed notification impossible to ever settle; settling is now a
status-free bookkeeping request, valid in any state. And the phase 2a guard
against shell references was matching hashlib.sha256 as a substring, so it now
requires a word boundary.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UrDNJkHmm18uu7n7JwV5Qd
The agent declares that its work is finished; the reader re-derives the hash
from the file rather than trusting the number in the declaration. A manifest the
watcher wrote itself would attest only that the watcher saw a stable file, which
is the self-certifying evidence shape rejected earlier in this issue.
Finalizing promotes a staged report by atomic rename before writing the
manifest, so a reader cannot observe a half-written report as final. Rewriting
the report afterwards invalidates the declaration immediately, because the
recomputed hash no longer matches -- a new body needs a new finalization.
A job that declared side effects must show evidence for every one of them.
Partial evidence is treated as none: the work is not known to be complete.
publish_event gains --finalize-report, --side-effect and --attempt-id, and does
the finalization before publishing. A `completed` event whose report is still
being written would invite exactly the partial adoption the contract forbids, so
a failure there fails the publish.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UrDNJkHmm18uu7n7JwV5Qd
Both reviewers sustained the round 1 fixes and found four more. Each was
reproduced here before being accepted; none were rebutted.
F-C (claude, blocking) -- my own round 1 fix caused this. Making refused writes
exit non-zero also classified an idempotent ACK as a refusal, and the wrapper
finalises an approved job with `status --set completed` on a record that is
already completed. It runs under `set -e`, so the *success* path aborted with
exit 4 before its validation hook and its final stdout line. Reproduced at the
shell. An idempotent result is now distinguishable from a refusal and exits 0; a
genuine conflict still exits 4.
F-D (claude) -- routing the operator write around update_status dropped the
status vocabulary check. Under the rollback flag an invalid status was written
to the record with exit 0, where before it was refused with exit 1 and the record
left alone. The check now runs first, in both modes.
R1 (codex, blocking) -- the previous fix compared identifiers only when they
happened to be present, so evidence carrying none at all was still approved on
the dispatch record's own flags. That is the self-assertion the check exists to
replace. Every identifier must now be supplied and agree, and a report manifest
must associate with the attempt being judged. Identifiers are read from a
collector envelope beside the payload rather than from inside it: a native Herdr
payload may not carry MAM fields, so one appearing there is either fabricated or
a contract violation, and either way it is not evidence.
R4 (codex) -- a sequence number orders nothing unless the compare and the write
are atomic. Two flushes could both read the old value, the newer land first, and
the slower overwrite it. Reproduced deterministically. The mirror now serialises
read-compare-write under a per-job audit lock, taken outside the registry
transaction so a slow mirror still cannot lengthen it, and each write uses its
own temp file instead of a shared name.
All eight new regressions fail against the pre-fix code.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UrDNJkHmm18uu7n7JwV5Qd
GodopuandClaude Opus 5