- tests: macOS SQLite keeps -wal/-shm after the last close (persistent WAL);
the `clean` state removes them after the checkpointing close
- the 2.x placeholder pattern requires the quoted example, so a typed
`Ask anything…` is read as a draft
- auth_ok takes ~/.opencode/bin/opencode only as an executable file
- status.sh and reconcile.sh (drift C) open opencode.db with _ro_connect
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
opencode 2.x needs no login: a fresh HOME runs a free default model and a
local provider has no key, while 2.x keeps no auth.json. auth_ok is now
"opencode on PATH or under <HOME>/.opencode/bin". `python -m lib_py.agents
auth-ok <agent>` exits with an adapter's verdict, and create's opencode
branch uses it (same 7 lines) instead of its own auth.json/key test and
`opencode --version`, which wrote a line to the real opencode log.
test_a4's opencode block now pins the binary rule (decision D1, required
by the brief's item 3 and by both reviewers). Other agents are unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
_ro_connect read every database without both sidecars as immutable, which
skips a -wal's committed rows: a recovered DB+WAL with no -shm showed no
sessions (planner-reviewer-codex-01). That state - no writer runs without
a -shm - is now read from a private DB+WAL copy, where SQLite may rebuild
the index, loaded into memory; nothing beside the source is touched. The
query matrix gains the walonly state.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
An agent that exits on /exit never turns `working`, so send_keys_safe
returns 4 although the stop then ends cleanly. The line now says the
delivery is unconfirmed and that the stop waits for the session to exit;
the kill steps keep their own lines, printed only when they run. Exit
judgement and the kill chain are unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every run appends a line to opencode's log, and one discover resolved the
database once per candidate. A successful answer is kept per (binary, HOME,
XDG_DATA_HOME) - the values the run sees; a failed run is not remembered.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A read-write open checkpointed the WAL and deleted -wal/-shm on close; a
plain mode=ro open created missing ones and left them. _ro_connect uses the
sidecars only when both exist (a live writer keeps them) and reads the file
as immutable otherwise. verify_artifact, _has_session, discover,
_descendants and residual_artifacts (db_rows_naming takes a connect) use it;
the purge's own writes are unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
An adapter can now name its placeholder as a pattern the box's first line
matches in full (input_placeholder_pattern); input_placeholder stays the
literal lib.sh greps for. opencode: `Ask anything… "<example>"`, the closing
quote optional because a box under ~65 columns cuts the example. Fixtures:
the 140-column herdr capture and a 60-column one with the cut example.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Review 8a485fa0 B1 / c9d06e85 / c4e96018 (blocking): the adapter now
follows the caller's XDG_DATA_HOME for its own home, but the tests did not
drop it: a sandbox that sets HOME resolved the runner's real data dir.
XDG_DATA_HOME joins _agent_home_guard.DROPPED_VARS (dropped per test), and
the session guard watches <XDG_DATA_HOME>/opencode when it is set.
Checked with XDG_DATA_HOME set to a separate dir: the opencode modules
pass and nothing is written there.
- status.sh asks the adapter for the db (`debug paths`, XDG-aware) instead
of a fixed ~/.local/share/opencode (8a485fa0 N1).
- The opt-in test's `opencode --version` probe ran with the real HOME at
import and appended to the real opencode log; it runs in a temporary home
now (8a485fa0 N2).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
- Review 614b6263 (codex, P2): `opencode session delete` removes the child
sessions with their parent, but their 1.x rows stayed and the parent's
residual was empty. The child sessions (session_v2 and a 1.x session
table, any depth) are taken before the delete; the 1.x tables (message,
part, todo, session_share, session) are cleaned for all of them, and the
residual also reports a child still on record (a failed delete).
- The in-use check counted any process holding the shared opencode.db; the
per-user 2.x background service (`serve --service`) holds it for every
session, so every opencode purge was refused while it ran (it does on
this Mac). It no longer counts; any other holder and a command line
naming the session still do.
- The data dir follows the caller's XDG_DATA_HOME when ctx is the caller's
own home (else <home>/.local/share), for debug paths, the fallback and the
residual scan (review 614b6263 / f4cedbae N4).
- resident_children = ('opencode',): the --standalone server stays the TUI's
idle child; what runs under it still counts (review 1abe0a90).
- Tests: child sessions on a migrated database, a failed delete, the service
holder, XDG, resident children, drift C pinning the parent over a newer
child session (review f4cedbae N2), and the real-binary test no longer
depends on PATH (N1).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
tests/_fake_opencode.py implements the two commands MAM calls (debug paths,
session delete) as measured on 2.0.32; unit tests never run a real opencode
(it would migrate the test database). test_real_opencode_2x_contract runs a
real 2.x opencode, when installed, in a home and tmp dir of its own with
--standalone.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
Measured on opencode 2.0.32 (macOS):
- the TUI rejects --agent ("Unrecognized flag"), so create and resume could
not start opencode at all; the default agent is build. MAM now spawns
`opencode --auto --standalone` (resume adds --session <id>): the server
runs as the TUI's child and ends with it, and the per-user background
service is neither used nor started;
- sessions live in session_v2 (a fresh 2.x database has no `session` table;
a migrated one keeps the 1.x tables but no longer writes them). discover,
verify_artifact, reconcile drift C and status read session_v2, and
discover / drift C skip child (subagent) sessions;
- `db path` is gone (it prints the help and exits 0): the db comes from
the `db` line of `opencode debug paths`, else the default path;
- `opencode session delete` removes a session with its child sessions from
the 2.x tables (and leaves the 1.x ones): purge runs it with --standalone
for a session in session_v2, still deletes the 1.x message/part/session
rows, and residual also reports a session_v2 row left by a failed delete.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
Review loop (--review-only, --all-reviewer, focused on this Mac:
macOS, /usr/bin/python3 3.9.6, bash 3.2): round 1 (@6f4f40a) agy and
grok PASS, claude and codex NOT PASS (a non-uuid id `memory` purged the
auto-memory dir, a linked dir below a root was followed, a NaN
startedAt counted as dead), sent as a PR comment and fixed by the
author in eleven commits; round 2 (@2a8d738) confirmed all three fixed
but 4/4 NOT PASS on one test holder (`sh -c "sleep 120" <uuid>` loses
the uuid on macOS /bin/sh), fixed in de6f3fc; round 3 (@de6f3fc) 4/4
PASS.
Full suite on macOS / Python 3.9.6 @de6f3fc: 1934 passed, 56 skipped,
1 error (the known #55 false positive: agy conversations of another
workspace). The real ~/.claude, /private/tmp/claude-<uid> and herdr
panes unchanged.
Also names lib_py.delegation.logroot in eager.CORE_MODULES (the v4.9.4
review's non-blocking note).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
macOS /bin/sh (bash 3.2) execs the lone `sleep 120` of `sh -c "sleep 120" <uuid>`,
so the uuid left the process list and both [upper] and [mixed] failed on the Mac.
A Python holder keeps the uuid in argv, like _holder() in test_issue10_13_purge.py.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
claude now takes only a canonical uuid (81b6c2e), so the fixture id
'uuid-c' purged nothing. The three claude spellings in
test_purge_artifacts_composite become one uuid literal; the assertions
and the other adapters' parts are unchanged. Approved by the user.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The user asked for no path at all from an id that is not a uuid. bd09c6a
still took `<id>.jsonl` for one (the pre-#31 reach), which has_artifacts and
the residual counted while in-use always said free: `purge claude notes`
deleted notes.jsonl. Now purge, has_artifacts and residual return nothing
for such an id, as in-use already did.
This makes test_a4_adapter_contract.py::test_purge_artifacts_composite fail:
it purges the claude id 'uuid-c' and expects uuid-c.jsonl gone. Editing an
existing test needs the user's approval; the one-line data change
(a canonical uuid in place of 'uuid-c') is kept outside the branch, in job
90efa4ce's report dir. Reverting this commit restores the jsonl-only reach.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
has_artifacts counted only the paths a purge may remove. When all that was
left lay behind a linked dir, `purge claude <uuid>` (or auto) found no
adapter and exited 5 with an empty residual, hiding the path the purge
had refused to follow. has_artifacts now counts every candidate; _inside
still decides what is removed, so the CLI exits 0, removes nothing and
lists the linked path as residual.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
_is_uuid took an uppercase or mixed-case uuid, and the paths built from it
are the live lowercase session's own files on macOS's default
case-insensitive APFS. The in-use checks compare exactly, though: session
files and command lines carry the lowercase id, so `purge claude <UPPER>`
found no holder and could remove a live session's files. _canonical now
lowercases a canonical uuid, and the paths, the session-file and job
matches, the in-use check and the residual all use it; a command line is
searched for both spellings.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
json.load reads a huge integer as an int, and math.isfinite raised
OverflowError on it outside any try: conversation_in_use and purge_artifacts
(which must never raise) threw, where 6f4f40a had taken it as live. The
conversion to float now comes first, and an overflow means live.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The refusal pointed at ~/.claude/sessions/ for every agent, whatever
CLAUDE_CONFIG_DIR said, and never mentioned a non-agent process naming the
uuid on its command line. The in-use CLI now prints, on stderr while the
conversation is in use, the adapter's in_use_hints: for every adapter
`ps -axww | grep <uuid>`, and for claude its real sessions/ dir first. stop
prints them under its refusal, and says "records preserved" instead of
"no state was modified": the gate runs after the kill chain, so the session
itself may already be down. SKILL.md's exit 4 row says the same; it is not
a code block, so no inventory hash moves.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
time.mktime read ps's local lstart, which is ambiguous in the hour a DST
change repeats: the start could come out an hour off and a live holder be
taken for a reused pid (or the reverse). ps now runs with TZ=UTC0, which BSD
ps and procps both honour, and calendar.timegm reads it. No /proc.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
float("NaN") and a JSON NaN passed the old check, and every comparison with
NaN is false, so a live holder looked like a reused pid: the in-use gate let
the purge through and the session file was taken as stale. startedAt must
now be an int or float (not bool), finite and positive; anything else cannot
be told and counts as live. A dead pid is still stale whatever it says.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A tmp-root entry <key> linking outside made `<key>/<uuid>` look like a plain
directory, and rmtree removed what lay behind the link; projects/<key>,
file-history/ and the other config dirs had no such check either. Each
candidate now keeps the root it was found under (config, projects or tmp),
and is purged only when every component between that root and the leaf is a
real directory and its parent resolves inside the resolved root. A root may
itself be a link (/tmp -> /private/tmp), a leaf link is still removed as the
link alone, and a candidate behind a link is reported as residual instead.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
projects/<key>/memory/ (Claude Code's auto memory) sits beside the session
dirs, and `purge claude memory` removed it whole. Only an 8-4-4-4-12 hex id
now reaches the session's directories, session files, jobs and tmp dirs. Any
other id keeps the pre-#31 reach, the `<id>.jsonl` file alone (test_a4 still
purges "uuid-c"); its residual is that file and nothing holds it in use.
The shared _plausible_id is unchanged for the other adapters.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The adapter's fixed /tmp fallback is now a module constant. The new tests
point it, TMPDIR, TMP and TEMP at the test's own dirs, in the test process
and (through a sitecustomize on PYTHONPATH) in every python a test starts,
so no test lists the real /tmp/claude-<uid> any more (Mac review, PR #59).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
End to end through stop_session.sh with CLAUDE_CONFIG_DIR alone and with a
matching CLAUDE_PROJECT_DIR; plus the DiscoveryContext default. The
config-only stop case and the default check fail on 872d5f1.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
stop resolves the uuid before it purges, through lib.sh's exported
CLAUDE_PROJECT_DIR and DiscoveryContext.claude_dir, both of which defaulted
to $HOME/.claude/projects. With CLAUDE_CONFIG_DIR alone the transcript under
<cfg>/projects was not found, so stop removed the row and purged nothing.
Both defaults are now ${CLAUDE_CONFIG_DIR:-<home>/.claude}/projects; an
explicit CLAUDE_PROJECT_DIR still wins.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Config-only and config+project overrides: a live session is refused and a
stopped one cleaned up. A key that cannot be removed keeps its session file;
a job dir that cannot be emptied keeps state.json; both are reported and
removed on retry. 5 of these fail on 8edb8d6.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
SKILL.md lines 52 and 83 still said a purge leaves a `terminated` row. The
two blocks' inventory shas move with them (#42), and so does the
passthrough line of stop's heredoc SystemExit(1) (310 -> 327, after the
process gate). Line counts are unchanged, so no other block moves.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A session file is what ties a pid's key file to the session, and a job's
state.json what ties the job dir: removing them first left the rest
unattributable after a partial failure (residual: none, and a retry could
not find it). Keys now go before their session file, which stays while any
key does; a job dir is emptied with state.json last and kept if anything
stays. residual lists a session file's key files (masked). A startedAt of 0
or less cannot tell a reused pid and counts as live; the sessions/ listdir
no longer raises.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
CLAUDE_CONFIG_DIR moves claude's whole config root, sessions/ included, so a
live session started under it was not seen and a stopped one left
file-history and the rest behind. The config root is now CLAUDE_CONFIG_DIR,
else <home>/.claude; the purge and residual scans look at ctx.claude_dir and
the root's own projects/, each once. The test guard drops the variable for
every test like the other agent-home variables (test_issue55 l1 requires
every *_DIR the adapters read to be accounted for).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
17 of the 25 cases fail on 26e2c43; the other 8 pin invariants (other
sessions, shared files, malformed ids, a prefix-twin job dir).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The prose said a purge leaves a `terminated` row; the code removes it, and
the monitor never marks a stopped row. The step-5 list names what a claude
purge removes. The two code blocks (lines 52, 83) still say `terminated`:
changing them changes their inventory sha, which waits for approval.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Their purge is unchanged; what it leaves is now reported: codex names and
thread rows (state_*, thread_history_*; logs_* is not scanned), grok
memory-v2 observations, opencode event/todo rows. db_rows_naming opens the
DB read-only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A closed herdr session is not an exited agent, and the `herdr already dead`
path ran no check at all, so stop could remove the files of a live claude
and then its row. New `python -m lib_py.agents in-use <agent> <uuid> [ws]
[--wait s]` gives the purge CLI's own verdict; stop asks it outside the
registry lock and, unless it answers free, exits 4 with nothing changed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
stop called purge_artifacts only. It now prints `WARN residual: <path>` for
each leftover or `residual: none (agent: <agent>)`, the strings orc_onboard
prints. The row is still removed and the exit stays 0.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
claude had no residual_artifacts, so a purge always printed `residual: none`.
It now reports names holding the uuid under ~/.claude (projects excluded),
in projects two levels deep, in the tmp roots, plus the owned jobs dir and
session files. Shared files (history.jsonl, ~/.claude.json) are not
reported: they always mention every session. scan_names goes to base.py.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The claude purge removed only <uuid>.jsonl. It now removes everything the
session alone owns: projects/<key>/<uuid>/, file-history/<uuid>/,
session-env/<uuid>/, jobs/<uuid[:8]>/ when its state.json names the session,
a stale sessions/<pid>.json with that pid's key file (name only, never read)
and <tmp>/claude-<uid>/*/<uuid>/. An id that is empty, all dots or holds
/, .. or * builds no path. has_artifacts finds leftovers without the transcript,
and conversation_in_use also sees a live claude through sessions/<pid>.json
(a `claude -c` names no uuid on its command line); a reused pid is told
apart by ps lstart against the file's startedAt.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Review loop (--review-only, --all-reviewer, focused on this Mac:
macOS, /usr/bin/python3 3.9.6, bash 3.2): round 1 (@63d8164) agy and
grok PASS, claude and codex NOT PASS (F1 paho via the /usr/bin/python3
fallback, F2 logs default path, R1 disk-fallback redispatch
interleaving, R2 symlinked registry), sent as a PR comment and fixed by
the author in four commits; round 2 (@a7cb90e) 4/4 PASS.
Full suite on macOS / Python 3.9.6 @a7cb90e: 1867 passed, 56 skipped,
1 error (the known #55 false positive: WAL/SHM of an agy conversation
running in another workspace). Real herdr panes and ~/.claude/projects
unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
A registry directory that is itself a symlink got two audit logs: submit
resolved --registry-dir with pwd -P and wrote next to the target, while logs
derived the sibling from the link path. sibling_logs_dir() now takes the
registry's realpath; an explicit DELEGATE_JOB_LOGS_DIR still wins.
Also: forwarded_keys() normalises both sides of its comparison, and the
design doc's status row lists the audit log next to the registry.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The record (status, token, attempt, transition_seq) was read before the
audit status.json; if A error -> redispatch (attempt 2) -> B completed
happened in between, the audit terminal was B's while the token and attempt
validated were A's, and A's caller was told "completed". The audit status
is now taken only when a fresh read of the record still shows the same
token, attempt and transition_seq; otherwise the job stays pending and the
next round reads both afresh. The no-record case is unchanged.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
cmd_logs always passed the wrapper's cwd-relative default registry, so the
usual run (registry not moved, cwd not the workspace) no longer found its
jobs. Without the option registry.py's own default decides, as before:
MAM_JOBS_DIR, then WORKSPACE_ROOT, then the cwd.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
control_plane and status.sh imported sibling_logs_dir from mqtt_common,
which imports paho at module level, so create/resume's binding and status
died on an interpreter without paho-mqtt (_delegate_py_bin's last fallback
is the PATH's python3). The registry/audit-log path rules move to
lib_py/delegation/logroot.py (standard library only) and mqtt_common
re-exports them. The package's own import chain also reached paho through
service -> publish_event -> publisher, so mqtt_common and publisher import
paho tolerantly and make_client() refuses with a clear error when a broker
is actually needed. control_plane, status.sh, lib.sh's delegate_publish_cmd
and delegate_logs_dir, and the wrapper's _logs_dir_abs import from logroot.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
New: prefix()/CLI forward three keys for a sibling binding and four for an
explicit root; create records the sibling on the row, omits it from the
pane's --env and still pins it in the injected publish commands; the
issue's scene (a pane with WORKSPACE_ROOT and MAM_JOBS_DIR only) moves its
audit log with --registry-dir and leaves the control root alone;
worktree_prepare's pane --env carries no DELEGATE_JOB_LOGS_DIR.
The three existing four-key forwarding assertions (user-approved) compare
FORWARDED_ENV = CP_ENV[:3]; CP_ENV and the negative assertions are kept.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
A worktree creator's binding always differs from its --workspace default,
so prefix() put DELEGATE_JOB_LOGS_DIR=<control root>/.mam/delegate_job_logs
into every such pane for good, and the env-first rule then kept an
experiment's --registry-dir from moving its audit log. forwarded_keys()
leaves that key out when it equals the sibling of mam_jobs_dir; a
non-sibling (explicit) root is forwarded as before. describe(), the forward
and whitespace checks and the row's four keys are unchanged.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
_check_disk_fallback converted transition_seq outside its try, so a
status.json whose sequence is not an integer would raise out of the
subscriber's main loop; such a status is now ignored with a warning, like a
stale one. delegate_started_recorded returned 1 before looking at the
registry's events.log when the audit-log root could not be resolved; it now
skips only the audit check.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The audit mirror is written after the registry commits, so a redispatched
record (attempt 2, pending) can sit next to attempt 1's completed
status.json, and a failed mirror leaves it there. The fallback validated the
token and attempt against the record, which say nothing about the audit
status's provenance, and closed attempt 2 with attempt 1's completion
(review P1, Issue #57). A versioned record (transition_seq) now takes the
audit status only when its transition_seq is at least as new; an audit
status without a sequence cannot prove which attempt it describes and is
ignored by a versioned record. An unversioned record (never transitioned,
or MAM_ENGINE_CAS=0) and a missing record keep the old rule.
Regressions: the redispatch/flush interval, the failed-mirror leftover, the
unversioned audit status, a newer audit-only terminal, and conflict
refusals and legacy claims next to a moved registry.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
engine._flush computed the registry's audit-log root but did not pass it to
the conflict_refused append, and _pick_pending_legacy did the same for its
status_changed append, so both landed in the workspace default when the
registry was moved and no override was set (review P2, Issue #57).
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
New file tests/test_issue57_logs_follow_registry.py: the resolver (L1),
registry.py (L2), the wrapper's submit/wait/logs (L3), publish_event.py
(L4), control_plane.from_env, worktree_prepare, status.sh, the comment-5
disk fallback, and the invariant that an explicit DELEGATE_JOB_LOGS_DIR
still wins. Every test ends by asserting the workspace's own
.mam/delegate_job_logs was never created.
Two existing tests gain monkeypatch.delenv("MAM_JOBS_DIR"): conftest's
autouse _isolate_mam_state_dirs gives every test a MAM_JOBS_DIR, and under
the sibling rule the WORKSPACE_ROOT/cwd fallbacks they assert would
otherwise resolve next to that registry. The order they assert (explicit
root, workspace, cwd) is unchanged.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
read_logged_status() returns the status.json dict; _check_disk_fallback
compared that dict with the terminal status strings, so a terminal status
recorded only in the audit log was never recognised (Issue #57, comment 5).
The dispatch-token and attempt checks are unchanged.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Both get_job_status implementations (JSON and text) try the
delegate_job_logs next to MAM_JOBS_DIR and the row's mam_jobs_dir after the
explicit log roots, so a job whose registry record was cleaned up still
shows its audit-only terminal status when the registry was moved.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
from_env() fills delegate_job_logs_dir with the delegate_job_logs next to
the resolved mam_jobs_dir unless DELEGATE_JOB_LOGS_DIR is set, so a session
created with only MAM_JOBS_DIR moved no longer records the workspace's old
log root. worktree_prepare.sh passes DELEGATE_JOB_LOGS_DIR on only when the
caller set it. lib.sh and the delegate-job wrapper no longer carry a bash
copy of the default: delegate_publish_cmd, delegate_started_recorded and
_logs_dir_abs resolve it with get_logs_dir(<registry>), the wrapper fails
closed when it cannot, exports the resolved root to its children, prints
it as submit's last line, and logs takes --registry-dir.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
get_logs_dir(registry_dir) resolves the audit-log root as the
delegate_job_logs next to the registry in use when DELEGATE_JOB_LOGS_DIR
is not set; sibling_logs_dir() is the pure sibling rule. Every writer and
reader that knows its registry (registry, engine, publish_event, remote,
job_subscriber, reconcile.sh, the loop runner's fingerprint exclusion)
passes that root instead of falling back to the workspace default.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Review loop (--review-only, --all-reviewer): round 1 (@3a30d7d) 4/4 PASS;
each reviewer reproduced W1 failing on the pre-fix tree. Seven #50/#55 and
leak modules (112 tests) from a clean worktree left no .mam in the
checkout.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
test_u1e_a_dry_run_record_is_not_an_open_job (PR #56) ran the delegate-job
wrapper without WORKSPACE_ROOT; the wrapper sources lib.sh, which then
defaults it to the checkout and rewrites <checkout>/.mam/shim/herdr
(found by review 38a64423). The node joins W1's list, which runs its
nodes in a copy of the checkout and fails on any .mam written there.
A scan of the #50/#55 and leak modules from a clean worktree found no
other writer.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
- exit 5: a busy or missing target cancels the job; a superseded attempt
leaves the record to its new owner, unchanged
- install.sh copies a legacy .env to .mam.env and keeps the original
- restore-off fallback: without tomllib the file keeps only that key
(restore off, other settings lost, warning); restore on only when no
file can be written
- PoC: no token only when none is passed; tests: every test gets a temp
registry and log dir, most also the sandbox HOME and mock herdr
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
- install.sh: what it actually checks; it writes a starter .mam.env with
MQTT_BROKER=localhost (broker.hivemq.com is only the unset fallback);
install_mam.sh is a separate installer.
- Security: tokens exist only when given or when the broker has TLS or a
username (PoC: no signature check); the remote dispatch carries the token.
- Server isolation: create defaults to a workspace-named server; the shim
bootstraps only a named server and never when MAM_NO_AUTOSTART=1
(status/monitor); restore-off is the normal path, with a warning fallback.
- Loop: the selected reviewers must all pass; no reviewer option means a
creator self-review.
- Interpreter order with the PATH python3 fallback; exit 5 also covers
target_not_found; two layers with their own registries; test isolation
stated as sandboxes plus after-run guards; quick start uses named sessions.
- README.ko.md now mirrors README.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
Rewritten against the current code: the per-workspace shim is
.mam/shim/herdr routing with --session (not /tmp/multi-agent-herdr-shim
with -L), registry writes run in the workspace .venv (no dual-interpreter
split), and the missing pieces are added - the delegation contract (ack,
heartbeat, budget, busy wait, remote transport, exit codes), the review
loop (strict verdicts, review identity, sub-jobs, report promotion),
create --onboard exits 44/45, the per-server pane claim, restore-off
servers, worktree helpers and the test isolation rules. Layout and quick
start follow the tree.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
Review loop (--review-only, --all-reviewer): round 1 (@325b056) 4/4 PASS,
round 2 (@db7f4ca, HERDR_SESSION) 4/4 PASS. Full suite at 325b056 on
macOS / Python 3.9 / bash 3.2 / herdr 0.9.3 from an agent pane:
1845 passed / 0 failed / 56 skipped (18 test_deploy_freshness skips: no
nats-docker submodule in the fresh worktree); the panes of the real
servers (default, multi-agent-mux) and ~/.claude/projects were unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
A suite run from an agent pane with --basetemp .../pytest-pr56-shim left
seven workspaces on that pane's real server (multi-agent-mux):
- the real-herdr resolvers (lib.sh _is_shim_path and the generated shim)
skipped every PATH directory with "-shim" anywhere in its path, so the
basetemp hid the mock's bin/ and the real herdr ran; only a directory
that is itself *-shim is a wrapper now;
- mam_sandbox kept HERDR_SOCKET_PATH (and the pane/tab ids), which the real
herdr follows over the sandbox HOME; it drops them now.
tests/test_herdr_real_server_leak.py: L1 (ancestor *-shim, both resolvers)
and L2 (pane routing dropped) fail before the fix; L1b keeps a *-shim
wrapper directory excluded.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
Merge review on macOS (--review-only loops, --all-reviewer): round 1 at
320aa9f 4/4 NOT PASS, round 2 codex F1 (a claimed dry-run job holds its
session), round 3 4/4 PASS (discarded: the worktree changed mid-review),
round 4 at 21f0c12 claude/agy/grok PASS with codex interrupted by a stream
disconnect, round 5 codex PASS. Full suite at 21f0c12 on macOS / Python
3.9 / bash 3.2 / herdr 0.9.3: 1860 passed / 0 failed / 38 skipped; no new
entries in ~/.claude/projects.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
A `default` scan runs without -L, so the shim answers it from the inherited
HERDR_SESSION_NAME; a create's reconcile --once labelled that server's panes
`default`, and since create adopts only a row of its own server (1e25cf2f
F2), parallel creates in one folder were refused (test_tier4_e2e scenario 4).
The skill-blocks data follows create_session.sh's SystemExit line.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
- admission: a `submit --dry-run` record is marked dry_run and is not an
open job of its session (review 1e25cf2f F1)
- create: the monitor-row adoption requires the row's server to be the one
this create spawned on (review 1e25cf2f F2)
- pane lock: retry only on contention (BlockingIOError); the docs say the
flock lasts while any fd-9 holder (the claim shell and the shim) lives
(review 66680d9d)
- tests on macOS: real-herdr box HOMEs under /tmp (sun_path is 104 bytes),
resolved paths, the fake claude reads non-canonical input (MAX_CANON 1024)
and the pane is read unwrapped; lock holders exec their sleep so killing
them frees the lock; the adoption tests name the server
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
The live onboarding test asserted that the first job's wrapper registered in
the same wall-clock second as the onboarding it had just placed at a second
boundary. Process start-up has no sub-second contract, so the setup assertion
failed now and then before any dispatch behaviour was checked. The test now
stamps the onboarding record, which it owns and which nothing else depends on
yet, with the job's created_at once that record exists; the stamp is written
by atomic replace so a scanning wrapper never skips a half-written record.
The lifecycle assertions and the fixed-id tie coverage in test_u2h are unchanged.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
open_jobs reports a live pending record nobody signals on as `injected`:
its producer registered and injected it without a waiter (create
--onboard, a job sent by hand), so it holds the session like a running or
admitted job, and a waiter of the same second with a lower job id, or one
registered a moment earlier and still in its preflight, no longer passes
it in the FIFO. The waiter signals on its own record before its first
scan, so its record is never mistaken for one.
lib_py.pane_lock resolves the shim's ambient target (no -L,
HERDR_SESSION_NAME empty or `default`) to the server of the pane the
caller runs in - HERDR_SESSION, else HERDR_SOCKET_PATH's layout, as
lib.sh _herdr_caller_server reads them - so an ambient and a named call
into one server claim under one lock; outside a pane it stays the
default server's.
Tests: u2h (the tie, the later-registered onboarding, the signalled
control), u1/u4 labels, c10 forces the same-second registration, p1b,
s2 (real herdr: held lock stops both call forms; mixed concurrent calls
take distinct panes). Two FIFO fixtures now carry a waiter's signal.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The registry wait judged its record by one rule: not this attempt's
pending record in this session -> superseded (redispatched, moved or
closed by someone else). That also caught the record's own `started`
landing before the prompt went out - the Issue #14 registry ack winning
over an unreachable broker, as test_issue30_stall_watch_stop acknowledges
at registration - and the wrapper left on exit 5 instead of its
broker_unavailable path (rc 3).
A record that is running under this attempt's token in this session is
this attempt's own progress: the waiter answers `clear` at once, writes
no admission and starts no hold. Another token, another session or a
terminal status stay `superseded` (review 55219be1 R1 unchanged).
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Three `create --onboard` at once into one folder (the control root, so no
control-plane prefix reaches the pane and the shim reuses an idle pane):
the shim's new-session chooses a pane - an idle one in a workspace on the
same cwd, a split, a fresh workspace's root pane - and then claims it with
`agent start`; two shims could choose the pane one of them had just made,
and the later start failed with agent_pane_busy (one create in three).
mam_herdr now runs every new-session call holding one lock per server: a
flock on fd 9, opened for the block and taken by lib_py.pane_lock, which
also bounds the wait (MAM_PANE_CLAIM_WAIT_S, default 120). It is released
with the block or with the process, however it ends. The shim heredoc is
unchanged (S11). Other servers' creates are not held.
Serialising the claims exposed a second race: a create's end-of-run
`reconcile.sh --once` adopted the pane of another create that had not yet
written its row, and that create refused its own name as already running
(exit 4). Such a row - registered by the monitor, no job, for a pane
created after this create began - is now replaced, not refused.
Tests: real herdr - shim-level and create-level three-way races into one
folder (the real herdr's pane-creating replies held back so the window is
hit on every run; both fail on the base with the orchestrator's error),
the different-folder test kept; pane_lock unit tests (per-server file,
timeout, kill -9 holder, handover, knob); the adoption rule and its three
refusals; shim contract h26 (held lock: no pane calls). The herdr box gets
a ~/.zshrc: without it every first `agent start` typed into zsh's first-run
menu and timed out at 30 s.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
R1: wait-open-jobs takes --dispatch-token (the token the wrapper read for
this attempt) and keeps it, with the session it asked about, for its whole
life; every write (signal, admission, the hold's renewal) is a guarded
engine merge that re-reads the record under the registry lock and refuses
once the record is not that attempt's pending one in that session. A record
redispatched, moved or closed by someone else makes the waiter answer
`superseded`: the bash wrapper exits 5 without a close, the record is its
new owner's. hold-admission is given the waiter's pinned identity instead
of reading a token from disk, so a hold that starts late holds nothing.
R2: a pending record that never carried a waiter's signal (create --onboard
registers and injects directly and waits ack_timeout_sec for `started`) is
live for that budget from created_at, not for the 30 s signal window; the
window applies only once a dispatcher has signalled.
Tests: u1 rows for the two pending lifetimes, u3b asserts the guard, u3f/u3g/u3h
(cross-session redispatch, in-place attempt change, token changed before the
waiter started) with the real waiter subprocess, u4e/u4f (late hold, running
hold released by a redispatch), s1 pins the token argument and the no-close
branch; real herdr C10 (onboarding valid before and after 30 s blocks the
session), C11 (expired onboarding is stale), C12 (redispatched waiter leaves
its record alone, exit 5).
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
F1: `clear` is now an admission persisted on the released record - a merge
through the engine whose new `guard` re-reads the session's open jobs under
the registry lock, so two waiters cannot both be admitted. It is scoped to
the attempt by the dispatch token (a redispatch voids it) and reported as
`admitted` to every other waiter, which holds the session as a running job
does, whatever its FIFO place: a lower id in the same second or a
redispatched older loop job now waits.
F2: a pending record's liveness window is the dispatcher's signal cadence
times three misses (10 s x 3 = 30 s), no longer ack_timeout_sec, which the
agent's budget takes down to 1 s. After `clear` a detached hold renews the
admitted record's signal until the record leaves pending, its attempt
changes, the wrapper is gone, or the wall clock ends.
Tests: u1 boundaries, u1d/u2e/u2f/u2g/u3d/u3e for the admission, u4c/u4d for
the hold, s1 pins --hold; real herdr: C3/C7/C8 adjusted, C9 is C8's
live-owner counterpart with short ack windows.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The `--status pending` reset is a redispatch and issues a new dispatch
token, but the loop read `_disp_tok` only after the registry and idle waits.
A busy or idle_wait close from phase two on therefore carried the previous
phase's token and the engine refused it, leaving the record pending while
the caller saw session_busy. The read now follows the reset directly; s1
pins reset < token read < registry wait.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
F1: from phase two on, the loop's `_registry_wait` ran before the record was
reset to pending, so the waiter held a terminal record: the engine refused its
dispatch_wait_at merge and other waiters never saw it in the FIFO. The wait
(and the idle wait after it) now follows the `--status pending` update; s1
pins the order and u3c shows why a terminal self record cannot signal.
F2: the real-herdr box teardown survives a hung stop/delete, kills a server
bootstrapped by a late client call, and asserts the server and HOME are gone.
F3: the delegate-job SKILL.md exit 5 row names the registry wait.
Also drops the dead loop in s1.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The FIFO is (created_at, job_id) at second resolution; B and C registered
within the same second once in the full suite, C sorted first by id and
cleared while B's signal was still fresh, so the stale line never printed.
C7 already keeps the 1.2 s gap; C8 now does too and asserts the order.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Merge-workspace review loop (--all-reviewer): round 1 (@a93c9bf) 4/4
PASS. Full suite at a93c9bf on macOS / Python 3.9 / herdr 0.9.3 with
WORKSPACE_ROOT set to an empty sentinel: 1752 passed / 0 failed / 38
skipped; the sentinel stayed empty, the review worktree had no .mam, and
the main checkout's shim was not touched.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
test_b13_reexec_preserves_original_argv and test_b13_no_freeze_switch_
disables_reexec ran the checkout's run_loop.sh with --creator, which gets
past argument parsing to mam_acquire_loop_lock on the default marker,
$MAM_REAL_ROOT/.mam/loop-guard-active with MAM_REAL_ROOT = the script's own
checkout. In the main checkout that is the live loop's lock (taken and
released for a moment: a real loop refused, or the test refused); in a review
tree it left an empty .mam. Both now set MAM_LOOP_MARKER under their tmp_path.
A sweep of the 16 modules that use run_loop.sh (a per-test watcher for a new
.mam in the checkout, an empty .mam removed before each test) found only
these two; the others pin MAM_LOOP_MARKER already, end before the lock
(--help, argument errors), or run a copy of the tree.
Tests: test_loop_marker_isolation.py L1 runs every test that reaches the lock
through the checkout's run_loop.sh in a copy of the checkout and requires no
.mam there; W1 of the Issue #17 isolation test now also requires that no .mam
exists (it only looked for files, so the empty lock directory passed it). At
97c2946 both fail.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGpJXhSZasJLU38ksDghe
Merge-workspace review loop (--all-reviewer): round 1 (@d198dd5) claude
agy PASS, grok codex NOT PASS (the same inherited WORKSPACE_ROOT in eleven
more modules; test_orc_onboard's import-time env); fixed in 8b47cc7.
Round 2 claude grok NOT PASS (test_o2_race_free_lock, test_o1_rebuttal
missed), agy codex PASS; fixed in ac48cb7. Round 3 (@ac48cb7) 4/4 PASS.
Full suite at ac48cb7 on macOS / Python 3.9 / herdr 0.9.3 with
WORKSPACE_ROOT set to an empty sentinel: 1751 passed / 0 failed / 38
skipped; the sentinel stayed empty and the main checkout's shim was not
touched.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
Two modules still ran the checkout's run_loop.sh - which sources lib.sh
before it parses an option - with the process environment and no root of
their own: test_o2_race_free_lock's test_o2_12 (review claude; it was in the
sweep's list and missed in 8b47cc7) and test_o1_rebuttal's
test_v3_rebut_option_parsing_and_budget, whose calls pass no env (review
grok; outside the env-copy grep). Both now use own_workspace_root, and W1
runs one node of each: at 8b47cc7 W1 fails with the copy's .mam/shim/herdr.
A last pass over the test files that run lib.sh, run_loop.sh, the wrapper or a
skill script: every one is marked, uses mam_sandbox or pins WORKSPACE_ROOT,
or does not source lib.sh (b7 diff_collect, deploy, issue42, issue45 config,
issue5_phase0, local_mqtt).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGpJXhSZasJLU38ksDghe
d198dd5 fixed one helper; the same defect was in eleven more modules
(reviews grok and codex R1/R2): their shells inherited the process
environment, so each `source lib.sh` rewrote <WORKSPACE_ROOT>/.mam/shim/herdr
- the checkout's own when the variable is unset or names this checkout (which
pytest_configure keeps), another workspace's under --noconftest.
- conftest: own_workspace_root, an opt-in fixture setting WORKSPACE_ROOT to
<tmp_path>/.mam-test-state/workspace. Not autouse: runner.py and
mqtt_common read WORKSPACE_ROOT in-process, and some tests unset it on
purpose. mam_sandbox and explicit envs run later and still win.
- pytestmark usefixtures("own_workspace_root") in test_version_consistency,
test_b19_headless_reconcile_fixes, test_b8_send_keys_verification,
test_issue6_fail_closed, test_loop_cli, test_issue5_phase3_loop,
test_issue5_phase4_runtime, test_o3_scoped_guard, test_issue25_plan_reuse,
test_orc_onboard (their shells inherit os.environ, many with no env at all).
- test_orc_onboard (R2): SCRUBBED_ENV is taken at import and no longer
carries WORKSPACE_ROOT; every call builds _scrubbed_env(root) with the root
of the call (ws / cwd / mam_sandbox, else this test's).
- test_workspace_scope (unittest) and test_issue21_worktree_prereqs._mutate
pin the root they already create.
Tests (test_issue17_workspace_root_isolation.py): W1 copies the checkout and
runs one test of each of those modules there with WORKSPACE_ROOT naming the
copy - the copy gets no .mam; W2 imports test_orc_onboard under a sentinel
root, then calls it under another - the sentinel stays empty. At d198dd5 both
fail (the copy gets .mam/shim/herdr; the sentinel gets it).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGpJXhSZasJLU38ksDghe
_env kept an inherited WORKSPACE_ROOT, so every helper that sources the
wrapper (lib.sh's _init_herdr_isolation) rewrote <that workspace>/.mam/shim/
herdr - another workspace's during the PR #51 review (v4.8.7 review, claude).
Only dropping it is not enough: lib.sh then derives the checkout itself, and
conftest keeps a WORKSPACE_ROOT that names this checkout, so the run would
rewrite the checkout's own shim (the real one when this is the main
checkout). _env now sets WORKSPACE_ROOT to the test's tmp dir.
Test: with a sentinel WORKSPACE_ROOT in the environment, _run_both and a
direct _env run leave the sentinel empty and the checkout's shim untouched,
and the shell sees the tmp dir. With the old _env the sentinel gets
.mam/shim/herdr.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGpJXhSZasJLU38ksDghe
Merge-workspace review loop (--all-reviewer): round 1 (@8241bd5) claude
grok PASS, agy NOT PASS (two static wrapper-literal tests did not follow
_decided/_exit_for), codex NOT PASS (R1: a yes/no decision that could not
run read as "no"); fixed in 541c3fe and 65b78b6. Round 2 (@65b78b6) 4/4
PASS. Full suite at 65b78b6 on macOS / Python 3.9 / herdr 0.9.3: 1730
passed / 0 failed / 56 skipped; 18 of the skips were the uninitialized
nats-docker submodule in the fresh worktree, and with it
tests/test_deploy_freshness.py is 31 passed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff
Godopu
