Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
32 changes: 32 additions & 0 deletions formulas/formulas_test.go
Original file line number Diff line number Diff line change
Expand Up @@ -228,6 +228,38 @@ func TestLibrarianStaleReviewBranchRecoveryStepHasNeedsInvestigationLabelAndGuar
}
}

// crn-4j0pa: "cairn stale-branches" reports status=error findings (a reviewer
// lookup failure, or ListReviewBranches itself erroring on that branch) with
// the same shape as escalate findings except .reviewer is always empty and
// .error carries the failure text (cmd/branches.go's evaluateBranch never
// sets both). Before this fix the step's jq filter selected only
// status=="escalate", so error findings -- arguably the ones most in need of
// a human's attention, since cairn itself couldn't even evaluate the branch
// -- were silently dropped every cycle instead of filed.
func TestLibrarianStaleReviewBranchRecoveryStepHandlesErrorStatus(t *testing.T) {
f := decodeFormula(t, "mol-cairn-librarian.formula.toml")
s, ok := stepByID(f, "stale-review-branch-recovery")
if !ok {
t.Fatal(`mol-cairn-librarian.formula.toml: no "stale-review-branch-recovery" step found`)
}

if !strings.Contains(s.Description, `select(.status == "escalate" or .status == "error")`) {
t.Error(`stale-review-branch-recovery: jq filter must widen to select(.status == "escalate" or .status == "error"), or error-status findings are silently dropped and never filed (crn-4j0pa)`)
}
if !strings.Contains(s.Description, `jq -r '.error // empty'`) {
t.Error(`stale-review-branch-recovery: must extract the finding's .error field (error findings have no .reviewer), or the filed bead's body cannot explain what cairn stale-branches itself failed on (crn-4j0pa)`)
}

notifyIdx := strings.Index(s.Description, `A reminder mail was already sent to`)
guardIdx := strings.Index(s.Description, `if [ -n "$REVIEWER" ]`)
if notifyIdx == -1 {
t.Fatal(`stale-review-branch-recovery: "A reminder mail was already sent to" sentence not found`)
}
if guardIdx == -1 || guardIdx > notifyIdx {
t.Error(`stale-review-branch-recovery: the "A reminder mail was already sent to $REVIEWER" sentence must be guarded behind a non-null-reviewer check (if [ -n "$REVIEWER" ]) placed before it, or error findings (no .reviewer) render "sent to " -- the crn-w4c6-shaped defect this must not reintroduce (crn-4j0pa)`)
}
}

// cairn cull-evict itself is not safely repeatable for the same entry --
// EvictToReviewBranch hard-errors once a proposal is already pending, since
// its review branch name is deterministic ("cull/" + entry ID; see
Expand Down
39 changes: 31 additions & 8 deletions formulas/mol-cairn-librarian-rig.formula.toml
Original file line number Diff line number Diff line change
Expand Up @@ -90,7 +90,8 @@ pass to pass; if it's instead re-cloned fresh every sweep tick, add
ENTRY_ID=$(printf '%s' "$finding" | jq -r '.entry_id')
TIER=$(printf '%s' "$finding" | jq -r '.tier')
AGE_SECONDS=$(printf '%s' "$finding" | jq -r '.age_seconds')
REVIEWER=$(printf '%s' "$finding" | jq -r '.reviewer')
REVIEWER=$(printf '%s' "$finding" | jq -r '.reviewer // empty')
ERROR_TEXT=$(printf '%s' "$finding" | jq -r '.error // empty')

EXISTING=$(bd list --label=dim:review-branch,source:cairn-librarian \
--title-contains="on $BRANCH (" --json --no-pager 2>/dev/null | jq -r '.[0].id // empty')
Expand All @@ -101,20 +102,42 @@ pass to pass; if it's instead re-cloned fresh every sweep tick, add
continue
fi

# crn-4j0pa: an error-status finding (cairn itself failed to resolve a
# reviewer, or ListReviewBranches errored) never carries .reviewer --
# cmd/branches.go's evaluateBranch sets exactly one of the two. Guard the
# reviewer-was-already-notified sentence on it instead of interpolating
# unconditionally, or this renders "sent to " for every error finding
# (the crn-w4c6-shaped defect).
if [ -n "$REVIEWER" ]; then
NOTIFY_LINE="A reminder mail was already sent to $REVIEWER on an earlier
cairn stale-branches pass at this same commit, and it is now past the
escalate threshold with no merge and no new commit since."
RESOLVER="$REVIEWER (or another reviewer)"
else
NOTIFY_LINE="cairn stale-branches could not evaluate this branch normally (see error below), so no automatic reviewer reminder was possible."
RESOLVER="a reviewer"
fi

ERROR_LINE=""
if [ -n "$ERROR_TEXT" ]; then
ERROR_LINE="

cairn stale-branches reported an error while evaluating this branch:
$ERROR_TEXT"
fi

NEW_ID=$(bd create \
--title="cairn librarian: stale review branch on $BRANCH ($TIER)" \
--stdin --type=task --priority=3 \
--labels=dim:review-branch,source:cairn-librarian,needs-pm,needs-investigation --silent <<BODY
Review branch "$BRANCH" (entry $ENTRY_ID, tier $TIER) has been unmerged for
${AGE_SECONDS}s. A reminder mail was already sent to $REVIEWER on an earlier
cairn stale-branches pass at this same commit, and it is now past the
escalate threshold with no merge and no new commit since.
${AGE_SECONDS}s. $NOTIFY_LINE$ERROR_LINE

Filed by the mol-cairn-librarian stale-review-branch-recovery step
(crn-0yv.1). This is a proposal only: it never merges the branch itself. To
resolve, $REVIEWER (or another reviewer) should review and merge the branch
into the store's default branch, or an author should update it -- either
action removes the branch from future "cairn stale-branches" output.
resolve, $RESOLVER should review and merge the branch into the store's
default branch, or an author should update it -- either action removes the
branch from future "cairn stale-branches" output.
BODY
)

Expand All @@ -130,7 +153,7 @@ BODY

echo "filed $NEW_ID for $BRANCH"
FILED=$((FILED + 1))
done < <(cairn stale-branches --store "$STORE" | jq -c '.[] | select(.status == "escalate")')
done < <(cairn stale-branches --store "$STORE" | jq -c '.[] | select(.status == "escalate" or .status == "error")')

echo "librarian stale-branch recovery: $FILED bead(s) filed, $SKIPPED already tracked"
'''
Expand Down
39 changes: 31 additions & 8 deletions formulas/mol-cairn-librarian.formula.toml
Original file line number Diff line number Diff line change
Expand Up @@ -99,7 +99,8 @@ pass to pass; if it's instead re-cloned fresh every sweep tick, add
ENTRY_ID=$(printf '%s' "$finding" | jq -r '.entry_id')
TIER=$(printf '%s' "$finding" | jq -r '.tier')
AGE_SECONDS=$(printf '%s' "$finding" | jq -r '.age_seconds')
REVIEWER=$(printf '%s' "$finding" | jq -r '.reviewer')
REVIEWER=$(printf '%s' "$finding" | jq -r '.reviewer // empty')
ERROR_TEXT=$(printf '%s' "$finding" | jq -r '.error // empty')

EXISTING=$(bd list --label=dim:review-branch,source:cairn-librarian \
--title-contains="on $BRANCH (" --json --no-pager 2>/dev/null | jq -r '.[0].id // empty')
Expand All @@ -110,20 +111,42 @@ pass to pass; if it's instead re-cloned fresh every sweep tick, add
continue
fi

# crn-4j0pa: an error-status finding (cairn itself failed to resolve a
# reviewer, or ListReviewBranches errored) never carries .reviewer --
# cmd/branches.go's evaluateBranch sets exactly one of the two. Guard the
# reviewer-was-already-notified sentence on it instead of interpolating
# unconditionally, or this renders "sent to " for every error finding
# (the crn-w4c6-shaped defect).
if [ -n "$REVIEWER" ]; then
NOTIFY_LINE="A reminder mail was already sent to $REVIEWER on an earlier
cairn stale-branches pass at this same commit, and it is now past the
escalate threshold with no merge and no new commit since."
RESOLVER="$REVIEWER (or another reviewer)"
else
NOTIFY_LINE="cairn stale-branches could not evaluate this branch normally (see error below), so no automatic reviewer reminder was possible."
RESOLVER="a reviewer"
fi

ERROR_LINE=""
if [ -n "$ERROR_TEXT" ]; then
ERROR_LINE="

cairn stale-branches reported an error while evaluating this branch:
$ERROR_TEXT"
fi

NEW_ID=$(bd create \
--title="cairn librarian: stale review branch on $BRANCH ($TIER)" \
--stdin --type=task --priority=3 \
--labels=dim:review-branch,source:cairn-librarian,needs-pm,needs-investigation --silent <<BODY
Review branch "$BRANCH" (entry $ENTRY_ID, tier $TIER) has been unmerged for
${AGE_SECONDS}s. A reminder mail was already sent to $REVIEWER on an earlier
cairn stale-branches pass at this same commit, and it is now past the
escalate threshold with no merge and no new commit since.
${AGE_SECONDS}s. $NOTIFY_LINE$ERROR_LINE

Filed by the mol-cairn-librarian stale-review-branch-recovery step
(crn-0yv.1). This is a proposal only: it never merges the branch itself. To
resolve, $REVIEWER (or another reviewer) should review and merge the branch
into the store's default branch, or an author should update it -- either
action removes the branch from future "cairn stale-branches" output.
resolve, $RESOLVER should review and merge the branch into the store's
default branch, or an author should update it -- either action removes the
branch from future "cairn stale-branches" output.
BODY
)

Expand All @@ -139,7 +162,7 @@ BODY

echo "filed $NEW_ID for $BRANCH"
FILED=$((FILED + 1))
done < <(cairn stale-branches --store "$STORE" | jq -c '.[] | select(.status == "escalate")')
done < <(cairn stale-branches --store "$STORE" | jq -c '.[] | select(.status == "escalate" or .status == "error")')

echo "librarian stale-branch recovery: $FILED bead(s) filed, $SKIPPED already tracked"
'''
Expand Down
77 changes: 77 additions & 0 deletions release-gates/crn-kq7tz-gate.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,77 @@
# Release Gate: crn-kq7tz

**Bead:** crn-kq7tz (deploy) — source: crn-lhchy (review), build: crn-4j0pa (plan: crn-zg69q)
**Reviewed commit:** `cacda2f2e3eff9455344abf81a33b4a59aaf17bc` (matches crn-lhchy's verdict and the deploy bead's `metadata.gc.deploy_commit` — see criterion 1)
**Deploy commit:** `cacda2f2e3eff9455344abf81a33b4a59aaf17bc` — no self-rebase needed, criterion 6 merge-tree came back clean directly
**Date:** 2026-08-20

## Background

`cairn stale-branches` reports two distinct finding shapes on the same
review-branch check: `status == "escalate"` (a reviewer exists but hasn't
acted) and `status == "error"` (cairn itself failed to evaluate the branch —
a reviewer-lookup failure, or `ListReviewBranches` erroring). `evaluateBranch`
sets exactly one of `.reviewer` / `.error`, never both. The
`stale-review-branch-recovery` step's jq filter selected only
`status=="escalate"`, so error-status findings — arguably the ones most
needing a human's attention, since cairn couldn't even evaluate the branch —
were silently dropped every sweep cycle instead of filed.

The fix widens the filter to `select(.status == "escalate" or .status ==
"error")` in both `formulas/mol-cairn-librarian.formula.toml` and
`formulas/mol-cairn-librarian-rig.formula.toml`, adds `ERROR_TEXT` extraction
(`.error // empty`) included in the filed bead body when present, and guards
the "reminder mail already sent to $REVIEWER" sentence behind a
non-null-`$REVIEWER` check (error findings never carry `.reviewer`) — avoiding
a crn-w4c6-shaped "sent to " defect on error findings.

## Criteria

| # | Criterion | Result | Evidence |
|---|-----------|--------|----------|
| 1 | Review PASS for exact deployed commit | PASS | crn-lhchy's verdict cites `cacda2f2e3eff9455344abf81a33b4a59aaf17bc`, identical to the deploy bead's `metadata.gc.deploy_commit` — exact SHA match (D = R). Independently re-resolved via `git rev-parse --verify --quiet "<sha>^{commit}"`, not trusted as a transcribed string. |
| 2 | Acceptance criteria met | PASS | Verified directly against crn-4j0pa's full `exit_contract` (5 bullets). Read the actual diff, not just builder/reviewer prose: confirmed jq filter widened in both formula files, `ERROR_TEXT` extraction present, `NOTIFY_LINE`/`RESOLVER` branching guards the reminder sentence on `[ -n "$REVIEWER" ]` in both files identically. Sequencing precondition crn-nc9te (6b45121, PR #126) independently confirmed reachable from origin/main via `git merge-base --is-ancestor` (also visible directly in `git log origin/main`). |
| 3 | Tests pass | PASS | Independently re-run at the deploy commit (detached checkout, clean tree). Pre-flight: this repo's suite has no container/testcontainers dependency (`grep -rl testcontainers\|DOCKER_HOST` empty), so the podman setup step is not applicable here. `go build ./...` exit 0; `go vet ./...` exit 0; `go test ./... -race -count=1` — **7/7 packages ok** (`.`, `cmd`, `formulas`, `internal/cairn`, `internal/critic`, `internal/obslog`, `scripts`), 0 FAIL, 0 SKIP. Diff-owned test mapped by exact name: `TestLibrarianStaleReviewBranchRecoveryStepHandlesErrorStatus` re-run individually (`-run '^...$' -v`) — PASS. 3a (attribute failing tests) not applicable, zero failures anywhere. 3b (policy/lint lane): this repo's only policy lane is the CI `lint` job itself (`golangci-lint run ./...` — confirmed via `.github/workflows/ci.yml`; no separate `ci-pr-policy`-style Makefile target exists here, `Makefile` targets are only `all/build/formulas/test/install/fmt/fmt-check/clean/help`). Run with an isolated `GOLANGCI_LINT_CACHE` (avoiding known cross-session cache contamination on this machine): `golangci-lint run ./...` — 0 issues; `golangci-lint fmt -d ./...` exit 0, no diff; `gofmt -l .` empty. |
| 4 | No open HIGH findings | PASS | crn-lhchy records **no blocking findings** ("No blocking findings. Handing off to deployer."). No open HIGH/blocking finding anywhere in the chain. |
| 5 | Clean tree | PASS | Verified via direct detached checkout of `cacda2f2e3eff9455344abf81a33b4a59aaf17bc^{commit}`; `git status --porcelain` empty. |
| 6 | Clean divergence from main | PASS | `git merge-tree --write-tree origin/main cacda2f2e3eff9455344abf81a33b4a59aaf17bc` → clean merge tree `4b1a3794db0e1c91c0d157a6b967b0f366ff5b49`, exit 0, zero conflicts — **no self-rebase needed**. `origin/main` has advanced by exactly one commit since this branch's fork point (`93c2593`) — the crn-bmy54 merge (`a7f9cdc`, PR #131) — and it does not touch either formula file this diff owns, so the merge stays conflict-free without rebasing. Pre-flight (has the target already merged?): `gh api repos/quad341/cairn/commits/cacda2f2e.../pulls` returned empty — no PR exists for this SHA — and `git merge-base --is-ancestor cacda2f2e... origin/main` confirmed it is NOT already on main. Normal flow applies. |
| 7 | Single feature theme | PASS | `git diff origin/main...cacda2f2e3eff9455344abf81a33b4a59aaf17bc --stat` (correct triple-dot form, symmetric-difference from the merge-base — a plain two-dot diff against current `origin/main` is misleading here since main advanced past the branch's fork point and a two-dot diff spuriously shows those advances "reverted") → exactly 3 files: `formulas/formulas_test.go` (+32), `formulas/mol-cairn-librarian-rig.formula.toml`, `formulas/mol-cairn-librarian.formula.toml` — 94 insertions, 16 deletions, all tightly related to the single fix. Out-of-scope paths (`packs/cairn-loop-orders/orders/cairn-librarian-rig-cooldown.toml`, `TestLibrarianStepsHaveNeedsPmLabelAndGuard`) confirmed untouched. |

## Verdict: PASS — proceeding to PR.

## Process notes

1. **Merge authority — self-merge authorized under the reinstated standing
authorization.** Same basis as `release-gates/crn-bmy54-gate.md` note 1:
mayor ruling `gm-wisp-2yhv7u` (2026-08-19) reinstated deployer self-merge
for quad341/cairn (not covered by `mpr`), and explicitly instructs future
sessions not to re-pause on a bare repeat of the generic "route to
mayor/mpr" boilerplate carried in this formula's own step text. crn-kq7tz's
notes raise nothing beyond that same recycled clause. Proceeding under the
STANDING AUTHORIZATION subject to the reinstatement's four conditions: (1)
gate 7/7 PASS — this table; (2) PR state confirmed via a direct, fresh `gh
pr view --json` read at merge time; (3) CLEAN/MERGEABLE with both required
checks (`build-test`, `lint`) COMPLETED/SUCCESS; (4) no `--auto` arm —
plain `gh pr merge <n> --squash --delete-branch` only (per mayor's
`gm-wisp-fvdoja` addendum), then independently verified.

2. **Pre-flight (target-already-merged check) run before criterion 6**, per
the formula's own ordering requirement (ga-xykek3 incident). No PR existed
for the deploy SHA and it is not yet an ancestor of `origin/main` — normal
flow, not reconciliation.

3. **SHA integrity:** the deploy commit was independently re-resolved via
`git rev-parse --verify --quiet "<sha>^{commit}"` on both the D (deploy
bead metadata) and R (review bead metadata) sides before comparison, never
trusted as an eyeballed or transcribed string.

4. **Two-dot vs. triple-dot diff pitfall (recorded for future gate
sessions):** a first pass at criterion 7 used `git diff origin/main
cacda2f2e...` (two-dot) and produced an alarming diffstat showing large
deletions across unrelated files (`internal/cairn/entry.go`,
`release-gates/crn-bmy54-gate.md`, etc.) — these were `origin/main`'s own
advances past this branch's fork point being spuriously shown as
"reverted," not real changes in this diff. Re-run with the triple-dot form
(`origin/main...cacda2f2e...`, symmetric difference from the merge-base)
immediately resolved to the correct, narrow 3-file diffstat matching the
reviewer's own count. Worth a formula-text callout if this recurs.
Loading