Once you synced a second repo, every repo synced before it showed a
false "run /sync-gbrain" prompt at session start. The read check now
uses the per-repo sync record `/sync-gbrain` already keeps.
Why: people who work across several repos saw this on every session.
Nothing to do after upgrading.
Co-authored-by: SatriaSuriaAriffin <57977752+SatriaSuriaAriffin@users.noreply.github.com>
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
gstack's Codex hooks no longer block every prompt when your Codex
folder's path has a `(` or `&`, and uninstall keeps your own Codex hooks
trusted.
Why: both broke Codex sessions; setup also now matches Codex's hook
rules, and the /review checklist on Codex reads shell commands more
carefully.
Nothing to do: plain paths keep the exact hook command they had, so
Codex asks for no new trust.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Review-Finding: p1/specialist-batched#1 reader=red-team
Review-Finding: lib/codex-rollout.ts:217:functional-contract reader=qa
Pre-commit /verify now fails a check that times out and runs its checks
on one time budget; a reviewer's scope-list bullets no longer need an
owner ruling to clear; stack tests fail when git can't list a PR's
tests.
Why: each let a result read clean with nothing proving it, or block on
nothing (#4518, #4519). Goal: gates never pass on missing evidence.
Nothing to do; `gstack-checks quick --timeout` is now one budget the
checks share (540 s default).
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Review-Finding: p1/outside-voice#1 reader=outside-voice
Review-Finding: p1/outside-voice#2 reader=outside-voice
/ship's second review round now reviews a fix that touched a gate-class
path, instead of reporting done with nothing read.
A plan told "nothing was fixed" while the branch has new commits since
the reviewed head now refuses with `review_required_no_readers` rather
than printing done.
Nothing to do: the next /ship picks it up.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Review-Finding: p1/red-team#1 reader=red-team
On Codex, gstack's /review can no longer end a session with a required
step skipped: the end-of-session check reads Codex's own session log and
sends the agent back to finish, as it already does on Claude Code.
Why: Codex users got routing and the destructive-command check in #4512,
but not the /review checklist, so a review could stop after the diff
read with no second-model pass.
To use it: run `./setup` again and trust the new hook in Codex
(`/hooks`); turn it off with `GSTACK_SKILL_CONTRACTS=off`.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Review-Finding: p1/specialist-batched#1 reader=testing
Review-Finding: p1/specialist-batched#2 reader=testing
When a second-model review fails, gstack now reports the failure that
actually ended it: Codex's final error decides, a stray "402" in a
request id no longer reads as a used-up budget, and a Claude Code 402 or
usage limit reads as a budget stop with no Codex advice.
Why: these were the four follow-ups ruled below the blocking bar on
#4515; each one could send someone to fix the wrong thing.
Nothing to do. The prose rule that reads "auth", "login" or "API key" in
stderr as a login failure is not changed here (see below).
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
On Capy, /ship's PR body edits, comments and new PRs no longer go out
through `gh` under the machine owner's GitHub account.
Each write runs its checks, saves the exact text to a private file and
prints a `HOST_PUBLISH:` line that names the repository (and host, off
github.com) it belongs to, so Capy's PR tool sends it to the same place
`gh` would have.
Nothing changes outside Capy (`GSTACK_CAPY` unset).
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Review-Finding: p1/specialist-batched#1 reader=specialist-batched
Review-Finding: p1/specialist-batched#3 reader=specialist-batched
Review-Finding: p1/red-team#6 reader=red-team
Review-Finding: p1/specialist-batched#1 reader=testing
Review-Finding: p1/specialist-batched#2 reader=testing
`./setup --host codex` now adds gstack's skill-routing hint and
destructive-command check to Codex's `~/.codex/hooks.json`, next to your
own hooks, so Codex users get the same guardrails Claude Code users
have.
Codex runs a hook only after you trust it, so nothing fires until you
accept gstack's hooks at Codex's startup prompt or in `/hooks`;
`gstack-doctor` shows whether you have.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Review-Finding: p1/red-team#1 reader=red-team
A repo can now list published-data folders under a new
`review_content_patterns` ship-policy key, so /ship leaves `.patch`
diffs, `MANIFEST.sha256` and extension-less licences out of review,
while source files there stay reviewed.
Why: gstack-evals #352 is 13.8 MB, 13.6 MB of it published session data,
and could never pass /ship's 2 MB review limit.
Nothing to do until a repo adds the key to its base-branch policy; a
pattern that keeps too many files now stops with
`REVIEW_EXCLUDE_KEEP_TOO_LARGE` instead of being dropped.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Review-Finding: p1/red-team#2 reader=red-team
Review-Finding: p2full/red-team#1 reader=red-team
When a pull request merges a release whose only extra change is a
generated file the repo lists under `release_outputs`, its earlier
remote test run now counts for it, if the clear command gets
`--allow-repo-commands`.
Why: gbrain paid one extra full CI run for each such pull request,
several a day.
Without the flag nothing changes except that the refusal now says to
pass it; a hand edit to the generated file, or any other changed file,
is still refused.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
On Codex, naming office-hours, plan-design-review or gstack-upgrade
loaded nothing, because setup left a second copy of each that Codex
listed twice; on macOS and Linux setup now links those copies, and
re-running setup or upgrading fixes an existing install.
Skills started under `codex exec` now report `SESSION_KIND: headless`
instead of `interactive` (also on slim Linux images without `ps`), Codex
reviewers start without the parent's conversation (`fork_turns:
"none"`), and the docs stop saying Codex has no hooks.
Nothing to do after merge; Codex users pick it up on their next upgrade.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Review-Finding: p1/red-team#1 reader=red-team
Review-Finding: p1/specialist-batched#1 reader=testing
When Claude Code calls /verify on its own right before `git commit`,
gstack now runs one quick check-only command that ends in a single
`VERIFY PASS`, `VERIFY FAIL` or `VERIFY skipped: ` line; typing /verify
still runs the full skill.
Why: since Claude Code 2.1.286 any skill named `verify` is run before
every commit, so gstack's fix loop, its first-run notices and a hard
stop in repos with no remote landed on ordinary commits.
Nothing to do; `gstack-config set verify_before_commit off` turns the
check off.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Review-Finding: p1/specialist-batched#1 reader=specialist-batched
Review-Finding: p1/specialist-batched#2 reader=specialist-batched
The prompt-routing hook can now suggest skills you installed yourself:
any skill that lists `triggers:` in its SKILL.md, not only gstack's
/review, /qa and /verify. It also stops sending ordinary bug reports to
/qa.
Why: a bug-fix skill you install was never named (0 of 60 real bug
reports); now it's named on 40 of 60, and /qa takes none of them instead
of 3.
Nothing to do after merge.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Review-Finding: p1/red-team#1 reader=red-team
Review-Finding: p1/specialist-batched#2 reader=api-contract
Review-Finding: p1/specialist-batched#1 reader=security
Claude Code now gets gstack's prompt-routing, skill-contract and
destructive-command hooks on by default, and new OpenClaw and Hermes
registrations get the start hook. On 60 real bug reports, routing now
matches a bug-fix skill on 40 (was 22), but gstack's own skills route 14
(was 17).
Why: on Opus 5.5, unattended, a hook line naming a skill got it loaded
in 58 of 60 sessions, and a skill relying on its description in 0 of 60
(gstack-evals#369).
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Review-Finding: p1v/outside-voice#1 reader=outside-voice
Review-Finding: p1v/adversarial-native#4 reader=adversarial-native
Review-Finding: p2/red-team#1 reader=red-team
/sync-gbrain's read check no longer reports `unknown` when the first
page in your code source is over 64 KB.
Why: the check read the whole page to confirm its source and slug, and
big pages (lockfiles) hit the 64 KiB output limit on every run.
Nothing to do: it now reads a 4 KB window of the page and still treats
any cut-off output as `unknown`.
Co-authored-by: nfitzsimmons <2198211+nfitzsimmons@users.noreply.github.com>
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Review-Finding: p1/specialist-batched#1 reader=testing
Commands that read large git or gh output, including the credential
pre-push hook, the pre-merge secret and literal scans and the
sealed-output guard, now fail when Bun cuts that output off, instead of
judging the part they got.
Why: Bun can report exit status 0 together with ENOBUFS and partial
stdout, so a push whose diff passed 64 MiB could get a credential past
the pre-push hook.
Nothing to do: the failure names `SPAWN_OUTPUT_INCOMPLETE` and links its
troubleshooting entry.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
`gstack-artifact validate` and `ack` now refuse a run.json that
gstack-autoplan never wrote, so a hand-built run can no longer pass as a
finished review.
Why: a session skipped `gstack-autoplan next`, built run.json with a
script copied from the reference fixture, and got ARTIFACT_VALID.
Nothing to do unless you validate old or hand-made run directories:
rerun those through gstack-autoplan.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
With a custom Codex provider, `gstack-codex-probe role-ready` and
`gstack-outside-voice run --runner codex-cli` now resolve the
plan-review model the same way, so a session no longer skips a review
that would have run.
Why: the check refused a `config.toml` naming `gpt-6-astra` while the
runner ran it, and a day-one eval session dropped the second-model
review in every phase.
Nothing to do: a custom provider naming the frontier catalog ID just
works; any other name now says how to opt in.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
A security finding that an owner ruled a note, dispositioned, or that
review deferred now shows in a PR body only as an opaque id and
"security finding, details recorded privately".
Why: on a public repository the old lines named the file, line and
reason of a live exposure outside the PR's change.
Nothing to do: other findings print exactly as before, and the details
stay in the private approval record.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Review-Finding: p1/specialist-batched#1 reader=security
Review-Finding: p1/red-team#1 reader=red-team
When /ship stopped on a second-model review finding marked "Blocks:
yes", the record it left on the PR said no blockers were found; now the
finding is counted and /ship will not write that record until it is
fixed or ruled.
Why: `ready` and the merge step trust that record, so an undercount let
a blocked PR look clean.
Nothing to do: the next /ship run logs Step 11's findings and checks
them before writing.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Review-Finding: h2p1/red-team#1 reader=red-team
Review-Finding: h2p1/red-team#2 reader=red-team
Review-Finding: h2p1/red-team#3 reader=red-team
Review-Finding: v1/outside-voice#1 reader=outside-voice
An unattended /autoplan run that's handed the owner's reply at the final
gate now records it and finishes, the same way whichever instruction
file the agent reads.
Before, autoplan/SKILL.md said "never approve" while autoplan/DRIVER.md
said to record the relayed reply, so the outcome depended on which file
the model followed.
Nothing to do; this fixes existing instructions and adds no new feature.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
An unattended /autoplan run on Claude Code now says its phase guard is
on, instead of saying it is off while the guard blocks it. A phase
report written only in reasoning now gets a denial that says so and
names the exact command to send.
Why: in the day-one eval, two runs were blocked by a guard their own
record called off, and one run ended blocked after four publish attempts
whose report was invisible.
Nothing to do.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
When /ship stops at a pull request's review round cap, a grant naming
that stop now buys one scoped review of the fix, in the same run or the
next, instead of another read-only full review.
Before, a grant recorded after the read-only review, or between /ship
runs, bought nothing, so teams paid for extra full reviews or second
grants (ghome#40: $8.78 against about $3).
Nothing to do: /ship prints the grant command when it stops.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Review-Finding: p1/specialist-batched#1 reader=security
Review-Finding: p1/specialist-batched#2 reader=state-machines
Review-Finding: p1/specialist-batched#3 reader=testing
Review-Finding: p1/red-team#1 reader=red-team
Review-Finding: p1/red-team#2 reader=red-team
Review-Finding: p1/red-team#3 reader=red-team
/ship's push step now always ends with one line that says what happened:
`PUSH: pushed …`, `PUSH: already_pushed …` or `PUSH: failed reason=`,
with the remote, branch and full commit.
Why: an eval grader looked for a `git push` call and counted 48 real
/ship pushes as "no push"; scripts and graders can now match this one
line instead.
If a script of yours matched the old `PUSH: ALREADY_PUSHED ` or `PUSH:
pushed to origin/` lines, update it; the exit status is unchanged.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
When /ship reuses a CI run that passed only after its failed jobs were
re-run, its CI line and its final summary line now add `rerun=
attempts=`.
Why: gbrain's CI now records which test files passed on a re-run
(gbrain#9690), and a pass on a re-run must read as flaky, never as
clean.
Nothing to do: runs without that record print as before; a malformed
record is refused with `CI_CARRY_RERUN_MALFORMED`.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Review-Finding: p1/outside-voice#1 reader=outside-voice
Review-Finding: p1/outside-voice#2 reader=outside-voice
Review-Finding: p1/adversarial-native#2 reader=adversarial-native
Review-Finding: p1/red-team#1 reader=red-team
A second-model plan review in `/autoplan` is no longer reported as
unavailable when it actually ran, and `gstack-outside-voice --help` no
longer prints a line that reads like a real consent refusal.
Why: sessions believed the false label and paid to re-run the review, or
closed the plan phase without its second-model review.
Nothing to do: the command autoplan already runs now picks the right
check by itself.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
/ship's review planner now sorts Swift and Xcode files by risk, so a
plain iOS UI change gets the light review instead of the full
five-reader panel.
Why: `gstack-diff-scope` had no rule for `.swift`, so every all-Swift
diff printed `SCOPE_ERROR=unmatched` and the planner failed closed.
Reported from the GHome iOS app's PR review.
Nothing to do: unknown file types still fail closed exactly as before.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Review-Finding: p1/outside-voice#1 reader=outside-voice
Review-Finding: p2/red-team#1 reader=red-team
Adds `gstack-first-use`: before an app build reaches its owner as "ready
to try", an agent from another model family plays that exact build, does
the plan's first five day-one actions with a screenshot or video of
each, and checks every link. Only then does it print the "try this
build" message.
Why: a recent app build passed every review and its owner still found
the first screen broken, because nothing tried the build the way the
owner would.
Nothing to do now; the first real use is that app's next build.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Review-Finding: p1/red-team#2 reader=red-team
Review-Finding: p1/red-team#4 reader=red-team
An agent can now check a draft question against the owner's earlier
answers before sending it, so the owner isn't asked again about
something they already said yes to.
Why: owners repeat "don't ask me, just do it" across threads, because
each new session forgets earlier answers. In a four-day sample, about 1%
of 444 asks were repeats this could skip, so a two-week pilot decides
whether it's kept.
Nothing to do: the new command is opt-in, and no existing command,
install output or factory rule changes.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Review-Finding: p1/specialist-batched#1 reader=testing
Review-Finding: p1/specialist-batched#2 reader=state-machines
Review-Finding: p1/red-team#1 reader=red-team
Review-Finding: p1/red-team#2 reader=red-team
Review-Finding: p3/outside-voice#1 reader=outside-voice
Review-Finding: p3/outside-voice#2 reader=outside-voice
/office-hours no longer carries its 34 founder videos and essays in the
skill file; the closing runs a new `gstack-founder-resources` command
that prints the ones a user hasn't been shown yet.
Why: the skill was 9 bytes from its size cap, and the approved
partner-session improvements need room. This frees about 8.6 KB.
Nothing to do as a reader.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Review-Finding: p1/specialist-batched#1 reader=testing
/ship's preflight and its CI-run lookup now reuse a passing CI test run
when a base merge left the files unchanged, stop instead of quietly
starting an hour-long test suite on the local machine, and refuse a
`--base` that names no commit instead of reporting an empty diff.
Why: gbrain #9662 lost a green CI run to a no-op base merge, then ran a
50-minute suite on a 4-CPU box, and its audit read a commit sha as a
branch and saw zero changed files.
Nothing to do: `--run-uncarried-here` restores running an uncovered
suite locally under `--tests-only`.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Review-Finding: s11/outside-voice#1 reader=outside-voice
Review-Finding: s11/outside-voice#2 reader=outside-voice
/sync-gbrain now says when an artifacts push failed or memory pages
failed to import, instead of printing OK and "pushed", and one malformed
queued path (from a skill started where the project slug was empty) no
longer stops artifacts sync for every project.
Why: people trusted those rows and lost track of unsynced work; fixes
#4445, #4418 and #4417.
Nothing to do after upgrading: a stuck `projects//timeline.jsonl` record
is dropped and reported on the next sync.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Review-Finding: p1/red-team#1 reader=red-team
Review-Finding: p1/red-team#2 reader=red-team
Review-Finding: p1/specialist-batched#1 reader=testing
Review-Finding: p1/specialist-batched#2 reader=testing
Review-Finding: p1/specialist-batched#3 reader=testing
Three small fixes: `skill_prefix=true` now survives updates when gbrain
is installed, the OpenClaw docs stop sending people to stale ClawHub
skills, and `gstack-issue-links` no longer reads other repositories'
issue numbers as gstack issues.
Why: each one misleads users or /ship today (wrong skill names after
every update, April 2026 skill builds, wrong `Refs #N` lines in PR
bodies).
Nothing to do as a reader. Thanks @fliaron (#4413) and @exGeni (#4421)
for the detailed reports.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
A driven /ship now runs each ship/DRIVER.md phase as one command, shows
how many calls the run made, and stops itself with a printed line at 100
minutes.
Why: measured /ship runs made 121 to 308 calls where the driver plans 9
to 12, and 10 of 12 ran past the 100-minute stop while those limits were
only prose.
Nothing to do: older ship records without `calls=` still validate and
print.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Review-Finding: p1/outside-voice#1 reader=outside-voice
* WIP plain PR bodies and squash messages
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Fold PR bodies to three plain lines and land every PR as one plainly titled commit
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Expect the plain commit body in the train land tests
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Count the new policy keys in the floor tests and record the ship size change
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Let older, generated and Dependabot PRs pass the body check, and keep comments posting
The opening check now follows the same date and branch exemption as the rest of the body check, Dependabot branches are exempt, the exemption only counts for branches in this repository, a merge that cannot read the PR waits instead of refusing, and a broken policy file no longer stops comments.
Review-Finding: p1/specialist-batched#1 reader=testing
Review-Finding: p1/specialist-batched#2 reader=testing
Review-Finding: p1/red-team#1 reader=red-team
Review-Finding: p1/red-team#2 reader=red-team
Review-Finding: p1/red-team#4 reader=red-team
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Stop re-posted PR bodies from duplicating sections, and let file names through the word check
Posting an already-folded body now replaces each section once instead of keeping the old copy beside it, and listed words inside file paths or file names, such as a test file in the Tests section, no longer count as prose.
Review-Finding: v2/outside-voice#3 reader=outside-voice
Review-Finding: v2/adversarial-native#2 reader=adversarial-native
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Describe the folded three-line PR body in the skills guide and the prose lint help
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
---------
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Keep checking a custom skill's contract when it still lists the removed test-after-edit item
The Stop hook used to treat such a contract as invalid and check none of
its items. It now skips only that item, warns with a line naming the file
and item, and checks the rest.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Move the changelog note to its own entry, since a landed entry is never edited
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* fix: pre-landing review fixes
Test that gstack-skill-contract check warns about a removed-kind item and checks the rest.
Review-Finding: p1/specialist-batched#1 reader=specialist-batched
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* chore: changelog and TODOS
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
---------
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Count review rounds per pull request across /ship runs (GST-2673)
/ship's two-round review cap was counted inside each run, so a pull request
shipped many times was reviewed again at every head. Step 1 now reads the
rounds the PR body's receipt history records, passes them to the review plan,
and sends later heads to the scoped re-check, with a lane or Captain round
grant for an extra round. Adds the no-loss replay preregistration.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Cap full review rounds per pull request and keep fix-checks to each run
The plan counts a PR's earlier full rounds (--prior-full) against the cap of
two, as the approved plan says, and prints every earlier round as pr_rounds=.
Renames the troubleshooting anchor to pr-round-cap, lists <SESSION_ID> as a
reviewed placeholder, and leaves ship/DRIVER.md at its size cap.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Point the ship driver's review pass at Step 9's plan flags, which carry the PR round count
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Give a pull request's final head a full review and make the diff-only re-check advice only (GST-2673)
A scoped re-check that reads only what changed since the last reviewed head
caught 22 of 24 planted faults but none of 8 real faults that full rounds had
caught, so it no longer carries a /ship receipt. A later head whose own lines
changed now gets a fresh full review, the PR's last before merge; a PR already
at two full rounds gets one read-only full review with no fix round, whose
blocker goes to the lane's Captain. carry prints RECHECK: advisory and refuses
--recheck. The drift threshold, the re-check blocker count and the
design-change grant kind are removed with the route they fed.
Repins lib/review-depth.ts and ship/sections/review-army.md under replay
GST-2673 (docs/evals/pr-round-count-results.md; deterministic grid against
main: 196,608 of 196,608 plans identical with no prior rounds).
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Keep a review round grant usable through its fix round and after a read-only review
A grant now holds after the fix round's commits move HEAD past it, counts only the rounds earlier heads spent, and a receipt written without a run keeps other runs' stubs.
Review-Finding: p1/specialist-batched#1 reader=state-machines
Review-Finding: p1/red-team#2 reader=red-team
Review-Finding: p1/red-team#3 reader=red-team
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Count every review round at a granted head again, so a grant can't reset the count
Reverts the history change that skipped rounds recorded at a round grant's head: the run that hit the cap is recorded at that same head, so skipping it wiped the count. A grant recorded after a read-only review still doesn't buy a fix round (fails closed).
Review-Finding: p2/adversarial-native#1 reader=adversarial-native
Review-Finding: p2/outside-voice#1 reader=outside-voice
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Give a merge head a full review when its removed lines changed or its receipt came from a re-check
A head whose removed lines changed with the base no longer carries on one scoped reader, a receipt an older gstack carried through a scoped re-check no longer carries on or lands, and a /ship body post keeps the run history. The read-only stop stops offering a round grant it can't honor, and the stamp-only-carry no-stop class is removed because a refused carry never stopped. (GST-2706)
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Keep the run history inside Agent notes when the PR text quotes a receipt
Review-Finding: p1/red-team#1 reader=red-team
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Let a receipt write skip a retired no-stop entry left in an older branch ledger
Review-Finding: p3/outside-voice#1 reader=outside-voice
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Say in troubleshooting that the queue refuses a receipt carried through a re-check
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
---------
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Stop /ship's docs audit halting when the child writes prose after its result (GST-2349)
The Step 14.5 documentation audit read only the child's last nonempty line.
Children that ended with their JSON result and then restated the documentation
section in prose had a valid "current, no edits" audit rejected with no format
repair, which spent an attempt each time and, after two, stopped /ship for an
owner risk grant (founderstar #837, harness-evals #393).
gstack-docs-candidate result now reads the last line that opens a JSON object,
ignores prose after it and says so with TRAILING_PROSE:. A result missing only
documentation_section keeps the existing free re-ask, whose prompt now tells the
child to put the section inside the JSON and write nothing after the line; the
re-audit retry note says the same. Content failures still count.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Send any docs-audit result with text after it back for one free re-ask (GST-2349)
A cross-family audit of the previous commit showed that ignoring prose after
the result line let a contradicting verdict through: a blocked result followed
by a fenced "current" example was accepted as current, and a "Correction: the
audit is BLOCKED" after a current line became a repair that had to stay current.
Text after a JSON line carrying the audit id is no longer ignored and never
searched for another result. It is a format slip: FORMAT_REPAIR re-asks once
for the whole result line without spending an attempt, the re-ask counts
everything the audit wrote after its JSON line, and the reply is judged on its
own with nothing kept from the first line. A reply that itself has text after
its line counts as an attempt.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Keep a blocked docs audit blocked through the free re-ask (GST-2349)
Review of the previous change found that the free whole-line re-ask could
turn a blocked audit into a passing one: a blocked result line, then a fenced
"current" example, then a "current" reply was accepted as current.
The re-ask now works one way. It is offered only when every JSON line that
carries the audit id is free of content problems; a content problem such as a
missing status or "current" with edits still spends an attempt. If any of those
lines said blocked or listed a blocker, the reply must stay blocked and keep
every blocker (it may add more), and the re-ask prompt says so.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
---------
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Make gstack's shell locks hold when mkdir reports a race it lost (GST-2247)
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Fix lock tests and fixtures for gstack_lock_take; trim setup-gbrain lock prose
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Ship the lock helper with every copied runtime and fixture that runs a lock
The Claude Code wrapper's copied runtime now includes bin/gstack-lock.sh, and
bin/gstack-render-claude.sh sources it beside itself, so gstack-config set
works in a copied runtime and the render lock is taken by every caller.
Test fixtures that copy single bin scripts now copy the helper too, and the
mkdir lock guard is registered as a repo-wide docs reader.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Release a lock whose owner file could not be fully written
gstack_lock_take created the owner file with a noclobber redirect and, when
that redirect failed, treated an existing file as a competing holder. If the
file was created but the write then failed (disk full, quota, ulimit -f), the
helper found its own empty file, returned 1 and left a lock nobody held;
freeze's state lock has no stale recovery, so it stayed busy until removed by
hand. The helper now tells "opened but write failed" from "already held",
removes its own file and directory, and returns 3. The inline copy in
gstack-ff-update is updated byte for byte, and a new test under ulimit -f 0
fails on the old helper.
Review-Finding: s11/outside-voice#1 reader=outside-voice
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Describe the lease lock as gstack_lock_take's O_EXCL owner file
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
---------
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Count review rounds per pull request across /ship runs (GST-2673)
/ship's two-round review cap was counted inside each run, so a pull request
shipped many times was reviewed again at every head. Step 1 now reads the
rounds the PR body's receipt history records, passes them to the review plan,
and sends later heads to the scoped re-check, with a lane or Captain round
grant for an extra round. Adds the no-loss replay preregistration.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Cap full review rounds per pull request and keep fix-checks to each run
The plan counts a PR's earlier full rounds (--prior-full) against the cap of
two, as the approved plan says, and prints every earlier round as pr_rounds=.
Renames the troubleshooting anchor to pr-round-cap, lists <SESSION_ID> as a
reviewed placeholder, and leaves ship/DRIVER.md at its size cap.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Point the ship driver's review pass at Step 9's plan flags, which carry the PR round count
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Give a pull request's final head a full review and make the diff-only re-check advice only (GST-2673)
A scoped re-check that reads only what changed since the last reviewed head
caught 22 of 24 planted faults but none of 8 real faults that full rounds had
caught, so it no longer carries a /ship receipt. A later head whose own lines
changed now gets a fresh full review, the PR's last before merge; a PR already
at two full rounds gets one read-only full review with no fix round, whose
blocker goes to the lane's Captain. carry prints RECHECK: advisory and refuses
--recheck. The drift threshold, the re-check blocker count and the
design-change grant kind are removed with the route they fed.
Repins lib/review-depth.ts and ship/sections/review-army.md under replay
GST-2673 (docs/evals/pr-round-count-results.md; deterministic grid against
main: 196,608 of 196,608 plans identical with no prior rounds).
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Keep a review round grant usable through its fix round and after a read-only review
A grant now holds after the fix round's commits move HEAD past it, counts only the rounds earlier heads spent, and a receipt written without a run keeps other runs' stubs.
Review-Finding: p1/specialist-batched#1 reader=state-machines
Review-Finding: p1/red-team#2 reader=red-team
Review-Finding: p1/red-team#3 reader=red-team
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Count every review round at a granted head again, so a grant can't reset the count
Reverts the history change that skipped rounds recorded at a round grant's head: the run that hit the cap is recorded at that same head, so skipping it wiped the count. A grant recorded after a read-only review still doesn't buy a fix round (fails closed).
Review-Finding: p2/adversarial-native#1 reader=adversarial-native
Review-Finding: p2/outside-voice#1 reader=outside-voice
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
---------
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Make /eval-plan check judges, zeros, failure counts and decisions before results count (GST-2754)
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Check zeros in every results table, apply Decisions and Scoring with stack: none, count only valid judge labels (GST-2754)
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Describe the /eval-plan result checks in the skills guide (GST-2754)
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
---------
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Garry, 2026-11-15: people rely on /plan-eng-review, so it returns as its own
command on every host that had it before GST-1698 (#4108). /autoplan's eng
phase is unchanged: external hosts still ship the same methodology beside
gstack-autoplan/SKILL.md, now rendered from the same pass as the standalone
skill, and the Kiro sibling copy and ownership guards from #4108 stay.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
gstack-redact refuses any input over 16 MiB before --max-bytes is read, so
a bigger --max-bytes never helped and nothing said so. --help and both
"too large" errors now name the 16 MiB limit, say --max-bytes can't raise
it, and suggest splitting the input. The help's exit line also says that
exit 3 covers input that is too large. What gets scanned, what gets
redacted, the exit codes and where output goes are unchanged.
Rebuilt on main from #4352.
Co-authored-by: ElNeMeR70 <83190847+ElNeMeR70@users.noreply.github.com>
Co-authored-by: exGeni <135786749+exGeni@users.noreply.github.com>
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
The standalone render now passes the requested role to the review-size ceiling, so the outside voice keeps its exemption there as it does in the pass manifest.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Reviewers state their blocker count in one line, and gstack checks it against the findings it parsed
Every blocking /ship reader's brief now asks for REVIEW_VERDICT: blocking=<n>. collect adds a CRITICAL row for each
declared blocker it did not parse, marks word-only CRITICAL rows as summary candidates when the count is lower, and
never lowers a row on its own except the native voice's unanchored FIXABLE raise. Ledgers rendered before this change
replay unchanged. Includes the preregistration for the no-loss replay.
Refs: GST-2495
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Add the verdict-line no-loss replay results
Refs: GST-2495
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Keep a finding line that mentions REVIEW_VERDICT instead of stripping it
Only a well-formed verdict line is removed before findings are parsed; prose that names the token, or a malformed attempt, stays in the text so its finding is still collected.
Review-Finding: p1/specialist-batched#1 reader=specialist-batched
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Check a stored summary flag against the reply on every ledger, not only ones that asked for the verdict line
verify now re-derives summary_candidate for every merged row, so a flag edited onto a real CRITICAL in an older ledger can no longer let judgment lower it while verify reports clean.
Review-Finding: v2/outside-voice#1 reader=outside-voice
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Never lower a finding automatically when a reviewer's blocker count is lower than the parse
A valid verdict line used to relabel the native voice's unanchored FIXABLE prose to its own label. Beside a
denial line such as 'No other CRITICAL issues.' that let a real blocker drop to FIXABLE while the count still
agreed. Those rows now stay CRITICAL and become summary candidates the lead lowers with a written reason.
Review-Finding: v3/adversarial-native#1 reader=adversarial-native
Refs: GST-2495
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Split a run-on sentence in the verdict-line changelog entry
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
---------
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Retry /ship's outside voice on Astra, then Gemini, when OpenAI refuses it under cyber_policy (GST-2649)
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Register the cyber_policy refusal fixtures as free test inputs (GST-2649)
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Retry a refused second-model review only when it left no review (GST-2649)
A voice the host marks refused under cyber_policy but that still wrote findings is now collected as that review, with its own outside record, instead of starting a redundant retry and losing its record. The cyber_policy test header now matches what the classifier does with a host error.
Review-Finding: p1/red-team#2 reader=red-team
Review-Finding: p1/specialist-batched#1 reader=testing
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Skip the Gemini retry when the lane or native reviewer is already a Google model (GST-2649)
On a Google-family lane, Step 11 refuses a Gemini outside voice, so two OpenAI refusals used to plan a Gemini retry that stopped the pass on every resume. The retry chain now skips a step whose family the lane or native reviewer holds and records the second-family read as missing.
Review-Finding: p1/outside-voice#1 reader=outside-voice
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Say in troubleshooting that a retry is skipped when its model family is the lane's (GST-2649)
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
---------
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Health checks find gbrain config under GBRAIN_HOME, trimmed the way gbrain trims it (GST-2625)
lib/checks.ts chose the gbrain health check from ~/.gbrain/config.json only, so under GBRAIN_HOME it judged the wrong brain. gbrainConfigDir() in lib/gbrain-exec.ts now follows gbrain configDir() exactly (trim(), refuse relative or .. values, return null), every caller treats null as no config, and checks.ts uses it. bin/gstack-gbrain-home.sh trims the JavaScript trim() set instead of POSIX [[:space:]], so a NBSP- or BOM-only GBRAIN_HOME means the default home. test/gbrain-home-parity.test.ts runs one table against both resolvers.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* gstack-gbrain-detect no longer crashes when GBRAIN_HOME is relative or has a .. segment
gbrainConfigDir() now returns null for a GBRAIN_HOME gbrain refuses. The detect script joined that value at load time and threw a TypeError before printing any JSON. It now treats a refused GBRAIN_HOME as no config, like the other callers, and the shape test covers it.
Review-Finding: p1/specialist-batched#1 reader=testing
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
---------
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Find gbrain's config under GBRAIN_HOME in the source wireup and skill start (#4325)
Both helpers read only ~/.gbrain/config.json. With GBRAIN_HOME naming one
brain and ~/.gbrain holding another, the wireup locked the ~/.gbrain brain's
database_url and sent sources add and sync there, and skill start skipped
its /sync-gbrain hint. Both now resolve the path through one shared helper,
bin/gstack-gbrain-home.sh, using gbrain's configDir() rule. The wireup gains
a read-only --check-target mode that tells an operator whether an earlier
run could have written to the wrong brain, without printing any URL.
Co-authored-by: exGeni <135786749+exGeni@users.noreply.github.com>
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Use localhost test database URLs the credential scan accepts (#4325)
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Test the shared gbrain config path helper directly (#4325)
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Report the caller's database URL settings in the wireup target check (#4325)
--check-target printed the URL env after the startup lock had exported GBRAIN_DATABASE_URL, so it said a URL was set when the caller had none. The helper comment no longer calls gbrainConfigDir() an exact twin.
Review-Finding: p1/specialist-batched#1 reader=red-team
Review-Finding: p1/specialist-batched#2 reader=testing
Review-Finding: p1/specialist-batched#3 reader=testing
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Document the GBRAIN_HOME config path and the wireup target check (#4325)
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
---------
Co-authored-by: exGeni <135786749+exGeni@users.noreply.github.com>
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Give a tests-only change one reviewer in the /ship review plan (GST-2547)
A diff that changed only test files planned zero readers, so the author's own core read was its only review. R51 gives tests one reviewer: the light runtime-0 plan now dispatches one batched Testing reader for a tests diff, and one batched code reader when the only code is registered outputs. Docs-only diffs still plan none.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* Repin the review planner after its no-loss replay (GST-2547)
Forced specialists on a tests diff keep their own reader, as on main; the batched Testing reader joins only when nothing else reads the diff. The deterministic no-loss replay over 186,624 planReview cases found 0 lighter plans and 2,392 heavier ones, each one added batched reader (replay GST-2547).
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
---------
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>