fix(runs,cli): refuse to remove a run dir under a live session (#526) by pbean · Pull Request #532 · bmad-code-org/bmad-loop

pbean · 2026-08-11T06:27:46Z

Implements both residuals of #526 (which carried #419's direction (3) past the #523 merge).

Residual A — the backstop

runs.delete_run / runs.archive_run removed a run directory with no session check. Their
docstrings said "callers enforce the live guard" and the callers do — but every caller's guard
is keyed on engine pid liveness, not session liveness: cli._stop_or_block_live_engine, the TUI's
action_delete_run / action_archive_run, and cmd_clean's retention path. An orphan — engine
dead, agent session still alive — is precisely the state that passes all three, and it is the one
state where the run dir is load-bearing: for an untagged session it is the only ownership proof
prunable_sessions can read, so removing it leaks the session (and its psmux server) for the life
of the machine.

Both removal paths now raise LiveSessionError while a matching session is live that this project
cannot prove belongs to another one. Per the observe-degrade / repair-raise doctrine:

runs.live_session_may_be_ours is an observation and degrades, each read in its own
direction. A listing that cannot answer reads as "no session" (what tmux already returns for a
dead server — a guard that fired differently per backend would be worse than none). A tag that
cannot be read is not proof the session is foreign, so it reads as untagged and the refusal
stands. Both MultiplexerError paths are caught explicitly: the seam only promises pipe_pane
and kill_session never raise.
the refusal is a repair-path write failing loudly, at the chokepoint below every caller.
ownership is read exactly as prunable_sessions reads it, so a session tagged to another project
— which carries its own proof and does not need this run dir — never blocks a removal.

Callers adapted, not widened:

delete / archive print the refusal and exit 1, appending the escape their own flag provides
(main's catch-all would give the same exit code but a bare error: …).
--force overrides the refusal and kills nothing. A session name carries no project, so
killing bmad-loop-<id> by name would tear down another project's live run on a --run-id
collision.
the refusal message names bmad-loop cleanup as the remedy together with the confirmation it
needs (bmad-loop attach <id>), because cleanup is not sound on its own here: for an untagged
session prunable_sessions proves ownership by that same run dir, so a shared run id prunes the
other project's session. No sound automatic remedy exists at this layer — that is Untagged sessions are weak ownership: they leak once their run dir is gone, and can be pruned by the wrong project on a run-id collision (the fallback behind #320) #419's thesis,
and minting a proof that outlives the run dir is its direction (2).
clean leaves such a run untouched, with the guard as the first thing in the reclaim loop —
ahead of reconcile_orphan_worktrees and trim_run_dir, not just ahead of the archive. If a
session appears mid-loop anyway, the chokepoint refusal is caught: the run is recorded and the
clean continues, classified by what actually happened (trimmed if the steps above already
reached it, protected only if nothing did — protected means "left untouched" in the --json
contract).
the TUI workers catch LiveSessionError alongside OSError so a refusal notifies instead of
killing the worker thread.

Residual B — the missing characterization test

test_prunable_sessions_claims_an_untagged_session_on_a_run_id_collision adds the case #419 named
and no fixture covered: untagged session, a dead run dir with the same id here, and the session
actually belonging to another project. Foreignness is constructed out of band — nothing in the
fixture marks it, because nothing can, which is the defect. Both projects' views are asserted, so
the test shows them disagreeing about one live session rather than echoing one side.

Pinned as the actual outcome, not the desired one: closing it needs an ownership proof that
outlives the run dir (#419 direction 2), not a change to either path in this PR.

Verification

Full suite 4971 passed / 39 skipped / 5 xfailed; uv run pyright 0 errors;
trunk check --all clean. CI 10/10 including both Windows jobs.

Every gate ablated singly, all bit (restored from file backups, never git checkout — that
reverts to the commit and silently drops in-flight fixes):

ablated	failing tests
`_refuse_live_session` calls in `delete_run`/`archive_run`	4 (2 runs-level, 2 CLI)
`force=args.force` pinned to `False` (override never reaches the leaf)	both `--force` tests
`force=args.force` pinned to `True` (guard never fires)	both CLI refusal tests
the `LiveSessionError` catch in `cmd_delete`/`cmd_archive`	both CLI refusal tests
the confirmation clause in the refusal message	the runs-level refusal test
the foreign-tag scoping (back to name-only matching)	the two foreign-tag tests
treating any tag as clearing the guard	the tagged-as-ours test
the `MultiplexerError` catch on the listing read	the listing-raises test
the `MultiplexerError` catch on the tag read	the tag-raises test
`cmd_clean`'s live-session branch	both `cmd_clean` guard tests (text + `--json`)
that branch moved back below `reconcile_orphan_worktrees` — position only, nothing deleted	the `cmd_clean` worktree assertion
the mid-clean `LiveSessionError` catch	the mid-clean race test
racer classification forced to `protected` / forced to `trimmed`	one race test each
`prunable_sessions`' untagged run-dir fallback	the new collision test (proves it is not vacuous)

Two ablations initially passed and the tests were strengthened rather than trusted: the CLI
refusal tests now assert (or pass --force) (only the handler emits it — main's catch-all
repeats the run message and would have hidden its removal), and the mid-clean race test now
asserts the untouched count (the else protected branch was unpinned).

Review rounds

Seven codex rounds (the last clean), plus three CodeRabbit passes. Codex found six real defects
after the first draft, the first two also raised independently by CodeRabbit:

--force killed the session by name → cross-project teardown on a run-id collision. Removed.
cmd_clean checked the session after reconcile_orphan_worktrees had already run. Guard moved
to the top of the loop.
The refusal recommended cleanup and the docstring called it "the scoped kill" — contradicted
by this PR's own collision test. Message and docs corrected.
The guard matched on session name alone, so a tagged foreign session blocked every removal
path, clean included, with no override. Scoped to accepted_tags.
A racing session made clean abort after partial mutation; and the run was then reported as
protected, which the --json contract documents as "left untouched". Both fixed.
MultiplexerError escaped the guard's observations on a conforming out-of-tree backend.

CodeRabbit then caught two documentation-consistency defects at head: two sites still described
cleanup as a scoped remedy (residue of correction 3), and clean_document's protected
description still said "past retention", which stopped being true when the guard moved to the top
of the loop in fix 2. Its one style finding — rewrite the CHANGELOG headline in the imperative —
was declined with a re-run count over the Unreleased section (54 entries open with "A", 22 with
"The", 7 with "An", 0 with an imperative verb); the verbosity half of that finding was taken and
the entry trimmed from 10 lines to 6.

Not covered

The TUI worker except widening has no test — app.py's lifecycle workers have no existing
harness, and standing one up for a widened except clause is out of proportion. The behavior it
guards is tested at the layer that produces it (runs.py) and at the CLI.
clean classifies runs once then mutates, so a concurrent resume can lose its worktree #533, filed from this review: clean classifies runs once and then mutates, so a concurrent
resume of a stopped run can lose its worktree. Identical on main at 814f0a2, where the loop
has no session check at all — this PR narrows the window but cannot close it, and closing it is a
concurrency-model decision (a lock, or documenting clean as unsafe alongside resume).
runs.mux_sessions()'s own docstring still assumes the bundled backend's "returns [] on
failure", and prunable_sessions reads it. Pre-existing, on an observation path rather than a
repair one, so the diff was not widened into it.

Closes #526. Refs #419, #523, #418, #320, #533.

`delete_run` / `archive_run` rmtree'd a run directory with no session check, and every caller's guard is keyed on engine pid liveness — so an orphan (engine dead, agent session still alive) passed all of them. For an untagged session the run dir is the only ownership proof `prunable_sessions` can read, so removing it leaked the session, and its server, for the life of the machine. Both removal paths now raise `LiveSessionError` while `bmad-loop-<run-id>` is live. `session_alive` degrades like every other observer here — an absent multiplexer, a dead server or a failed query reads as no session — so an unanswerable question never blocks a removal the operator is entitled to. `delete`/`archive` surface the refusal with the escape their own flag provides; `--force` kills the session first, satisfying the guard rather than bypassing it. `clean` classifies such a run as protected instead of removing it. Also adds #419's named missing test: an untagged session whose run id collides with a dead run dir in another project is claimed by that project, while its own project reads it as live. Pinned as the actual outcome — closing it needs an ownership proof that outlives the run dir, not a change to either path here.

coderabbitai · 2026-08-11T06:27:59Z

Walkthrough

Run removal now checks matching agent-session liveness. delete and archive refuse live sessions, while clean preserves protected runs. --force bypasses the refusal without terminating the session. CLI, TUI, documentation, and regression tests cover this behavior.

Changes

Live session protection

Layer / File(s)	Summary
Session liveness guard `src/bmad_loop/runs.py`, `tests/test_runs.py`	Added `LiveSessionError` and live-session ownership checks. Deletion and archival refuse potentially owned live sessions. Tests cover ownership, exact matching, multiplexer failures, collisions, and forced behavior.
Command and cleanup integration `src/bmad_loop/cli.py`, `tests/test_cleanup.py`, `tests/test_cli.py`	CLI commands report live-session conflicts. `clean` preserves protected runs, handles sessions that appear during cleanup, and reports partial reclamation. Forced removal does not terminate the session.
TUI and documented behavior `src/bmad_loop/tui/app.py`, `src/bmad_loop/documents.py`, `docs/FEATURES.md`, `CHANGELOG.md`	TUI workers report live-session conflicts. Documentation describes protection, cleanup guidance, protected results, and force behavior.

Estimated code review effort: 4 (Complex) | ~45 minutes

Sequence Diagram(s)

sequenceDiagram
  participant Operator
  participant CLI
  participant RunStore
  participant Multiplexer

  Operator->>CLI: delete, archive, or clean run
  CLI->>RunStore: request removal or cleanup
  RunStore->>Multiplexer: check bmad-loop-run-id session
  Multiplexer-->>RunStore: live, foreign, or unavailable result
  RunStore-->>CLI: protected result, LiveSessionError, or removal result
  CLI-->>Operator: report outcome

Possibly related issues

clean classifies runs once then mutates, so a concurrent resume can lose its worktree #533: The issue modifies cmd_clean and adds live-session checks for run reclamation, which matches this cleanup behavior.

Possibly related PRs

bmad-code-org/bmad-loop#518: Both changes modify runs.py and multiplexer session handling.
bmad-code-org/bmad-loop#522: Both changes modify multiplexer session detection and error handling.
bmad-code-org/bmad-loop#523: Both changes modify project/session ownership and run-ID collision handling.

Suggested reviewers: dracic

Poem

A rabbit guards each living run,
Until its session work is done.
Clean leaves protected paths in place,
Force removes without session chase.
Tests thump softly in the sun.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 55.26% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	The changes address both residuals in issue `#526`: live-session removal protection and untagged run-ID collision test coverage.
Out of Scope Changes check	✅ Passed	The implementation, documentation, and tests remain focused on live-session protection, forced removal, cleanup races, and collision coverage.
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and concisely describes the main change: preventing removal of run directories while a live session exists.

✨ Finishing Touches 💡 1

📝 Generate docstrings 💡

Create stacked PR
Commit on current branch

🧪 Generate unit tests (beta)

Create PR with unit tests
Commit unit tests in branch fix/526-live-session-run-dir-backstop

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

_{Comment @coderabbitai help to get the list of available commands.}

pbean · 2026-08-11T06:28:00Z

@codex review

pbean · 2026-08-11T06:33:01Z

@coderabbitai full review

coderabbitai · 2026-08-11T06:33:32Z

✅ Action performed

Full review finished.

chatgpt-codex-connector

💡 Codex Review

bmad-loop/src/bmad_loop/cli.py

Line 3071 in 0b0dbf1

runs.delete_run(run_dir)

Catch late live-session refusals in clean

If the first multiplexer observation returns no session because of a transient query failure, or a session starts after that sample, the second check inside delete_run/archive_run raises LiveSessionError here. cmd_clean does not catch it, so clean --json exits with only an stderr error instead of its required JSON object; catch the backstop refusal around both removal calls and classify the run as protected.

AGENTS.md reference: AGENTS.md:L31-L31

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

coderabbitai

Actionable comments posted: 2

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@src/bmad_loop/cli.py`:
- Around line 3054-3066: Move the runs.session_alive(run_dir.name) guard to the
beginning of the reclaimable loop, before runs.reconcile_orphan_worktrees() and
artifact trimming, so live sessions are added to protected and skipped
immediately. Add a regression test covering a live agent session with a
populated worktree and verify the worktree is preserved.
- Around line 2883-2889: The force-removal path around
`runs.kill_session(run_id)` must require durable proof that the session belongs
to the current project; do not terminate untagged or otherwise unowned sessions
by `run_id` alone. When ownership cannot be established, refuse the
delete/archive operation, and add CLI coverage for two projects claiming the
same untagged session ID to ensure the other project’s live session remains
intact.

🪄 Autofix

Fix all unresolved CodeRabbit comments on this PR:

Push a commit to this branch (recommended)
Create a new PR with the fixes

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: a90708f4-cf47-4158-816f-b7d1a6793110

📥 Commits

Reviewing files that changed from the base of the PR and between 814f0a2 and 0b0dbf1.

📒 Files selected for processing (9)

CHANGELOG.md
docs/FEATURES.md
src/bmad_loop/cli.py
src/bmad_loop/documents.py
src/bmad_loop/runs.py
src/bmad_loop/tui/app.py
tests/test_cleanup.py
tests/test_cli.py
tests/test_runs.py

…e every mutation Two P1s from the review gate, both raised independently by codex and CodeRabbit. `--force` killed the session by name before removing. A session name carries no project, so on a `--run-id` collision — reachable, the flag is caller-supplied and validated for shape only — `delete --force` in one project tore down another project's live run. This turned the guard's own false refusal into destructive cross-project teardown. `--force` now overrides the refusal and kills nothing; the scoped kill is `bmad-loop cleanup`, which proves ownership from the tag first, and the refusal message already names it. `cmd_clean` checked the session inside the past-retention branch, after `reconcile_orphan_worktrees` had already force-removed the worktree the live session may still be working in — and after the trim path, which a run inside retention reaches with no check at all. The run was then reported "protected" having already been half-reclaimed. The guard is now the first thing in the loop, ahead of every mutation. Its test grows a real worktree and moves past retention so the ordering is what fails: relocating the guard back below the reconcile (position only, nothing deleted) reddens it.

pbean · 2026-08-11T06:50:39Z

@codex review

pbean · 2026-08-11T06:50:40Z

@coderabbitai full review

coderabbitai · 2026-08-11T06:50:45Z

⚠️ Action not completed

Review rate limited.

Your included review limit is currently reached under our Fair Usage Limits Policy. This review may still proceed through usage-based billing if eligible. Your next included review will be available in 42 minutes.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 14f8c7c019

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

Codex, round 2: the message told the operator to clear the session with `bmad-loop cleanup`, and the docstring called that the scoped kill because `prune_sessions` reads the tag. Both are wrong for the exact state the guard fires on. `prunable_sessions` proves ownership from the tag only when there IS one; for an untagged session it falls back to this same run dir, so on a shared run id project A's `cleanup` prunes project B's live session — the edge this PR's own collision test asserts. My rationale contradicted my own test. No sound automatic remedy exists here — that is #419's point, and minting an ownership proof that outlives the run dir is its direction (2), not this guard. So the message keeps `cleanup` but carries the confirmation it needs (`bmad-loop attach <id>`) and states plainly why. Pinned by the refusal test; ablating the confirmation out of the message reddens it. Same correction applied to the CHANGELOG entry and FEATURES.md, which repeated the claim.

pbean · 2026-08-11T07:04:39Z

@codex review

pbean · 2026-08-11T07:04:40Z

@coderabbitai full review

coderabbitai · 2026-08-11T07:04:46Z

⚠️ Action not completed

Review rate limited.

Your included review limit is currently reached under our Fair Usage Limits Policy. This review may still proceed through usage-based billing if eligible. Your next included review will be available in 28 minutes.

chatgpt-codex-connector · 2026-08-11T07:08:46Z

Codex Review: Didn't find any major issues. Already looking forward to the next diff.

Reviewed commit: ff0f876593

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

pbean · 2026-08-11T07:14:46Z

@coderabbitai full review

coderabbitai · 2026-08-11T07:14:54Z

⚠️ Action not completed

Review rate limited.

Your included review limit is currently reached under our Fair Usage Limits Policy. This review may still proceed through usage-based billing if eligible. Your next included review will be available in 18 minutes.

pbean · 2026-08-11T07:25:35Z

@coderabbitai full review

coderabbitai · 2026-08-11T07:25:41Z

⚠️ Action not completed

Review rate limited.

Your included review limit is currently reached under our Fair Usage Limits Policy. This review may still proceed through usage-based billing if eligible. Your next included review will be available in 7 minutes.

pbean · 2026-08-11T07:33:40Z

@coderabbitai full review

coderabbitai · 2026-08-11T07:33:47Z

✅ Action performed

Full review finished.

coderabbitai

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@CHANGELOG.md`:
- Around line 191-200: Replace the verbose changelog narrative with one short,
scannable, imperative release-note item summarizing the run-directory removal
protection. Remove the collision rationale and multiplexer behavior from this
entry, leaving those details to the feature documentation.

🪄 Autofix

Fix all unresolved CodeRabbit comments on this PR:

Push a commit to this branch (recommended)
Create a new PR with the fixes

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 207e13dd-4a2b-4b49-8252-daa4bc671933

📥 Commits

Reviewing files that changed from the base of the PR and between 814f0a2 and ff0f876.

📒 Files selected for processing (9)

CHANGELOG.md
docs/FEATURES.md
src/bmad_loop/cli.py
src/bmad_loop/documents.py
src/bmad_loop/runs.py
src/bmad_loop/tui/app.py
tests/test_cleanup.py
tests/test_cli.py
tests/test_runs.py

…e length

pbean · 2026-08-11T07:38:59Z

@codex review

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 25ad3794d5

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

Codex P2: the guard matched on session name alone, so project A's delete, archive, TUI actions and every `clean` reclamation were blocked whenever a caller-supplied run id collided with project B's *tagged* live session — until B's run exited, and `clean` has no override to escape with. A tag outside `accepted_tags` proves the session foreign, and a tagged session carries its own ownership proof: it does not need this project's run dir, so removing the dir strands nothing. The guard now reads ownership exactly as `prunable_sessions` does and stands down there. Untagged, or tagged as ours, still refuses — an unread tag is not proof of anything, and only the untagged case was ever load-bearing. `session_alive` becomes `live_session_may_be_ours(project, run_id)`, which `delete_run` needs the project for; it now takes one, matching `archive_run`. The session listing is read before the tag query, so the extra call only happens on a name collision. Both directions ablate: name-only matching reddens the two foreign-tag tests, and treating any tag as clearing the guard reddens the tagged-as-ours test.

pbean · 2026-08-11T07:47:43Z

@codex review

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: a9284904a7

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

… the clean Codex round 4. The loop-top guard is one sample, so a resume racing this clean can start a session after it and before the removal. The chokepoint refuses there — correctly — but nothing caught it, so one racing run took the whole invocation down through main's catch-all, after the trim had already run. The removal branch now catches LiveSessionError, records the run as protected, corrects the freed estimate down to what the trim actually reclaimed, and carries on with the remaining runs. Ablating the catch reddens the new test, which flips the ownership read between the two calls — a stateless fake would answer the same both times and pin nothing. The wider race the finding describes is left alone and filed as #533: every mutation in that loop runs against a `reclaimable` sample taken in the previous loop and never re-read, so a resumed run can lose its worktree — identically on main before this PR, where the loop has no session check at all. Closing it is a concurrency-model decision (a lock, or documenting clean as unsafe alongside resume), not a guard fix.

pbean · 2026-08-11T07:56:16Z

@codex review

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: c8fb69165a

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

…ntended Codex round 5. A run the race caught after `reconcile_orphan_worktrees` and `trim_run_dir` had already taken its artifacts was still recorded as `protected` — the field `clean --json` documents as "left untouched", and the count text mode prints as untouched. A consumer would read a partially reclaimed run as a preserved one. It is now classified by the state it ends in: trimmed when anything above reached it, protected only when nothing did. Both branches ablate — forcing either one reddens one of the two race tests, which cover the two sides.

pbean · 2026-08-11T08:02:37Z

@codex review

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 27a804e735

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

…kend Codex round 6. `live_session_may_be_ours` claimed both its reads degrade to "no answer", but that is the bundled tmux backend's behavior, not the seam's contract: only `pipe_pane` and `kill_session` are contractually best-effort, so a conforming out-of-tree backend raises MultiplexerError from `list_sessions` and `session_options`. That exception escaped through `delete`, `archive`, `clean` and the TUI workers, turning a transient transport error into a failed removal — and `clean` has no override. Both reads are now caught, each degrading in its own direction. A listing that cannot answer reads as "no session", matching what tmux already returns for a dead server — a guard that fired differently per backend would be worse than none. A tag that cannot be read is not proof the session is foreign, and by then the listing has established a session is live, so it reads as untagged and the refusal stands. Each `except` ablates against its own test.

pbean · 2026-08-11T08:09:18Z

@codex review

chatgpt-codex-connector · 2026-08-11T08:12:53Z

Codex Review: Didn't find any major issues. Chef's kiss.

Reviewed commit: 7809e9d448

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

pbean · 2026-08-11T08:13:36Z

@coderabbitai full review

coderabbitai · 2026-08-11T08:13:42Z

⚠️ Action not completed

Review rate limited.

Your included review limit is currently reached under our Fair Usage Limits Policy. This review may still proceed through usage-based billing if eligible. Your next included review will be available in 20 minutes.

pbean · 2026-08-11T08:33:33Z

@coderabbitai full review

coderabbitai · 2026-08-11T08:33:44Z

⚠️ Action not completed

Review rate limited.

Your included review limit is currently reached under our Fair Usage Limits Policy. This review may still proceed through usage-based billing if eligible. Your next included review will be available in 8 seconds.

pbean · 2026-08-11T08:34:22Z

@coderabbitai full review

coderabbitai · 2026-08-11T08:34:27Z

✅ Action performed

Full review finished.

coderabbitai

Actionable comments posted: 2

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@src/bmad_loop/cli.py`:
- Around line 3026-3041: Qualify the cleanup comment near
runs.live_session_may_be_ours in src/bmad_loop/cli.py:3026-3041 to state that a
later clean reclaims the run only after the operator confirms ownership with
bmad-loop attach <id>. Update the prune_sessions docstring in
tests/test_cli.py:2142-2159 to state that untagged sessions fall back to the
local run directory, which is not sufficient ownership evidence across projects
with identical run IDs.

In `@src/bmad_loop/documents.py`:
- Around line 438-443: Update the documentation text around the protected-run
description in documents.py to state that protected runs may be inside or beyond
the retention window when they carry a live agent session, while preserving the
existing explanations for --keep-listed and non-terminal runs.

🪄 Autofix

Fix all unresolved CodeRabbit comments on this PR:

Push a commit to this branch (recommended)
Create a new PR with the fixes

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 053956ab-580d-48ad-8e3f-b75d3f9d069e

📥 Commits

Reviewing files that changed from the base of the PR and between 814f0a2 and 7809e9d.

📒 Files selected for processing (9)

CHANGELOG.md
docs/FEATURES.md
src/bmad_loop/cli.py
src/bmad_loop/documents.py
src/bmad_loop/runs.py
src/bmad_loop/tui/app.py
tests/test_cleanup.py
tests/test_cli.py
tests/test_runs.py

CodeRabbit at head, two Minor findings, both accurate. Round 3 corrected the claim that `bmad-loop cleanup` is a scoped remedy — it is not, for the untagged case the guard fires on — in the refusal message, the `_refuse_live_session` docstring, FEATURES.md and the CHANGELOG. Two sites still carried the old wording: the comment on `cmd_clean`'s guard and the docstring of the --force test. Both now say what the message says: `cleanup` clears the session, but ownership is confirmed first. `clean_document`'s description of `protected` said "past retention but still carrying a live agent session". The guard sits at the top of the reclaim loop, so a run *inside* the window is protected the same way; a consumer modelling `protected` off that sentence would be wrong about which runs can appear.

pbean · 2026-08-11T08:46:54Z

@codex review

chatgpt-codex-connector · 2026-08-11T08:50:15Z

Codex Review: Didn't find any major issues. What shall we delve into next?

Reviewed commit: 09aa7fa3aa

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

pbean · 2026-08-11T08:50:57Z

@coderabbitai full review

coderabbitai · 2026-08-11T08:51:04Z

⚠️ Action not completed

Review rate limited.

Your included review limit is currently reached under our Fair Usage Limits Policy. This review may still proceed through usage-based billing if eligible. Your next included review will be available in 43 minutes.

chatgpt-codex-connector Bot reviewed Aug 11, 2026

View reviewed changes

Comment thread src/bmad_loop/cli.py Outdated

Comment thread src/bmad_loop/cli.py Outdated

coderabbitai Bot reviewed Aug 11, 2026

View reviewed changes

Comment thread src/bmad_loop/cli.py Outdated

Comment thread src/bmad_loop/cli.py Outdated

chatgpt-codex-connector Bot reviewed Aug 11, 2026

View reviewed changes

Comment thread src/bmad_loop/runs.py Outdated

coderabbitai Bot reviewed Aug 11, 2026

View reviewed changes

Comment thread CHANGELOG.md Outdated

docs(changelog): trim the run-dir removal entry to the section's hous…

25ad379

…e length

chatgpt-codex-connector Bot reviewed Aug 11, 2026

View reviewed changes

Comment thread src/bmad_loop/runs.py Outdated

chatgpt-codex-connector Bot reviewed Aug 11, 2026

View reviewed changes

Comment thread src/bmad_loop/cli.py

pbean mentioned this pull request Aug 11, 2026

clean classifies runs once then mutates, so a concurrent resume can lose its worktree #533

Open

chatgpt-codex-connector Bot reviewed Aug 11, 2026

View reviewed changes

Comment thread src/bmad_loop/cli.py Outdated

chatgpt-codex-connector Bot reviewed Aug 11, 2026

View reviewed changes

Comment thread src/bmad_loop/runs.py Outdated

coderabbitai Bot reviewed Aug 11, 2026

View reviewed changes

Comment thread src/bmad_loop/cli.py

Comment thread src/bmad_loop/documents.py Outdated

pbean merged commit cfde647 into main Aug 11, 2026
10 checks passed

This was referenced Aug 11, 2026

fix(adapters,engine): tell a lost mux session apart from an exited CLI #522

Merged

Untagged sessions still have no ownership proof that outlives the run dir (#419 direction 2, now unhomed) #538

Open

coderabbitai Bot mentioned this pull request Aug 11, 2026

test: cover envvars registry + TUI hard-stop/archive paths; fix inf session-timeout override #541

Merged

Uh oh!

Conversation

pbean commented Aug 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Residual A — the backstop

Residual B — the missing characterization test

Verification

Review rounds

Not covered

Uh oh!

coderabbitai Bot commented Aug 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Walkthrough

Changes

Sequence Diagram(s)

❌ Failed checks (1 warning)

Uh oh!

pbean commented Aug 11, 2026

Uh oh!

pbean commented Aug 11, 2026

Uh oh!

coderabbitai Bot commented Aug 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

Uh oh!

coderabbitai Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

pbean commented Aug 11, 2026

Uh oh!

pbean commented Aug 11, 2026

Uh oh!

coderabbitai Bot commented Aug 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

pbean commented Aug 11, 2026

Uh oh!

pbean commented Aug 11, 2026

Uh oh!

coderabbitai Bot commented Aug 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

chatgpt-codex-connector Bot commented Aug 11, 2026

Uh oh!

pbean commented Aug 11, 2026

Uh oh!

coderabbitai Bot commented Aug 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pbean commented Aug 11, 2026

Uh oh!

coderabbitai Bot commented Aug 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pbean commented Aug 11, 2026

Uh oh!

coderabbitai Bot commented Aug 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

coderabbitai Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

pbean commented Aug 11, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

pbean commented Aug 11, 2026 •

edited

Loading

coderabbitai Bot commented Aug 11, 2026 •

edited

Loading

coderabbitai Bot commented Aug 11, 2026 •

edited

Loading

coderabbitai Bot commented Aug 11, 2026 •

edited

Loading

coderabbitai Bot commented Aug 11, 2026 •

edited

Loading

coderabbitai Bot commented Aug 11, 2026 •

edited

Loading

coderabbitai Bot commented Aug 11, 2026 •

edited

Loading

coderabbitai Bot commented Aug 11, 2026 •

edited

Loading

coderabbitai Bot commented Aug 11, 2026 •

edited

Loading

coderabbitai Bot commented Aug 11, 2026 •

edited

Loading

coderabbitai Bot commented Aug 11, 2026 •

edited

Loading

coderabbitai Bot commented Aug 11, 2026 •

edited

Loading