shard(tick): 0554z-otto-cli-2026-05-16

docs/BACKLOG.md

@@ -679,5 +679,6 @@ are closed (status: closed in frontmatter)._
 - [ ] **[B-0536](backlog/P3/B-0536-orphan-ferry-ref-cleanup-and-audit-false-positives-2026-05-15.md)** Orphan-ferry-ref cleanup + audit false-positive on filename paths
 - [ ] **[B-0537](backlog/P3/B-0537-memory-md-index-entry-lengths-cleanup-and-gate-2026-05-15.md)** memory/MEMORY.md index-entry length cleanup (100 long entries) + CI gate
 - [ ] **[B-0553](backlog/P3/B-0553-audit-backlog-status-drift-detection-2026-05-16.md)** Backlog status-drift auditor — detect `status: open` rows whose primary artifact has already shipped
+- [ ] **[B-0557](backlog/P3/B-0557-audit-backlog-status-drift-quality-improvements-2026-05-16.md)** Audit-backlog-status-drift — quality improvements per PR #3758 reviewer findings
 
 <!-- END AUTO-GENERATED -->

docs/backlog/P3/B-0557-audit-backlog-status-drift-quality-improvements-2026-05-16.md

@@ -0,0 +1,78 @@
+---
+id: B-0557
+priority: P3
+status: open
+title: "Audit-backlog-status-drift — quality improvements per PR #3758 reviewer findings"
+tier: factory-hygiene
+effort: S
+created: 2026-05-16
+last_updated: 2026-05-16
+depends_on: [B-0553]
+composes_with: [B-0553]
+tags: [audit, hygiene, follow-up, reviewer-findings, mechanization]
+type: feature
+---
+
+# Audit-backlog-status-drift — quality improvements
+
+## Origin
+
+Follow-up to [B-0553](B-0553-audit-backlog-status-drift-detection-2026-05-16.md). PR #3758 (the audit-tool implementation) attracted 4 additional reviewer findings on the e5e7143 commit (after the regex-order + SKIP_SECTIONS + nested-headings fixes). Each finding is valid; this row captures them as a follow-up slice rather than continuing the iteration treadmill on the original PR.
+
+## The four findings
+
+### Finding 1 — Mixed-bullet path extraction (Codex P1, PR #3758 line 172)
+
+> Do not skip the entire line when an inline cross-reference token appears.
+
+Current behaviour: `INLINE_CROSSREF_PATTERNS.some((re) => re.test(line))` skips the WHOLE line. This is correct for pure cross-reference bullets like `- Composes with \`tools/x.ts\`` but is overly aggressive for mixed bullets like `- Add \`tools/new.ts\` per [B-0123] convention` — the path `tools/new.ts` is a deliverable, the `per [B-0123]` is just a citation.
+
+**Proposed fix**: only skip the line if the cross-ref keyword appears at the START of the bullet (after `^[\s*-]+\s+`). Mid-line cross-refs allow extraction.
+
+### Finding 2 — cwd-relative path resolution (Copilot P1, PR #3758 line 212)
+
+> Tool assumes cwd = repo root. `enumerateOpenRows()` reads `docs/backlog/...` and `findDriftCandidates()` calls `existsSync(p)` on relative paths. If invoked from another cwd, the paths fail to resolve.
+
+**Proposed fix**: detect repo root via `git rev-parse --show-toplevel` (or accept it as a CLI flag with default), then resolve all paths absolute from that root. Use `path.resolve(repoRoot, p)` in `existsSync` calls.
+
+### Finding 3 — Error handling on read failures (Copilot P1, PR #3758 line 202)
+
+> `enumerateOpenRows()` can throw and abort the whole run on a single unreadable/edge-case backlog file (e.g., permission denied, transient FS error).
+
+**Proposed fix**: wrap `readFileSync` in `try/catch`; on failure, emit a stderr warning naming the failed file and skip that row (continue with the next). The full audit completes; the operator sees the partial-result warning.
+
+### Finding 4 — `--check` mode for CI integration (Copilot P1, PR #3758 line 277)
+
+> B-0553 spec calls for a CI-friendly "check/enforce" mode (non-zero exit when drift candidates are found). `main()` currently always exits 0.
+
+**Proposed fix**: add `--check` flag. When set, `main()` returns 0 if no candidates, non-zero (e.g., 65) if any candidates. Default behaviour (detect-only) unchanged.
+
+## Acceptance
+
+- Mixed-bullet path extraction: extracts deliverables from bullets that contain cross-ref tokens mid-line; skips only bullets that START with cross-ref keywords. Regression tests cover both cases.
+- cwd-independent path resolution: tool runs correctly from any directory; verified via `cd /tmp && bun /path/to/tools/hygiene/audit-backlog-status-drift.ts`.
+- Error handling: tool emits stderr warning + continues on unreadable files; verified via mock fixture with a permission-denied file.
+- `--check` mode: returns non-zero when candidates present; default mode unchanged. CI wire-up can later add this flag to a GitHub Actions cron.
+- 16-test baseline preserved; ≥3 new regression tests for the new behaviour.
+
+## Composes with
+
+- [B-0553](B-0553-audit-backlog-status-drift-detection-2026-05-16.md) — parent row (audit-tool spec + first-slice impl)
+- PR #3758 reviewer findings (verbatim references above)
+- [`.claude/rules/backlog-item-start-gate.md`](../../../.claude/rules/backlog-item-start-gate.md) step 0 — the discipline the audit tool mechanizes
+- [`memory/feedback_substrate_drift_catch_pattern_*.md`](../../../memory/) — the discipline substrate
+
+## Why P3
+
+Same as B-0553 — friction-reducer. Each improvement is small and additive; together they take the tool from "first slice" to "operationally robust." Not P2 because the first slice already works for the common case (30+ live candidates detected correctly). Not P4 because the findings are concrete + actionable + recent.
+
+## Non-goals
+
+- `--prune-claims` flag (still deferred per B-0553)
+- `--open-close-pr` flag (still deferred per B-0553)
+- Cross-row drift detection (per B-0532 — different audit)
+- Auto-merge integration (separate concern; the tool is a detector, not an actor)
+
+## Origin tick
+
+Tick 15 of the 2026-05-16 session. PR #3758 review-cycle 2 produced these 4 findings on commit `e5e7143`; this row captures them as follow-up substrate so PR #3758 can ship cleanly.

docs/hygiene-history/ticks/2026/05/16/0554Z.md

@@ -0,0 +1,68 @@
+# Tick 2026-05-16T05:54Z — Otto-CLI
+
+Sixteenth tick of the resume-session series. Brief-ack #3 of new
+counter cycle. Rate limit at 0/5000; **reset in 1 min** — sweep
+deferred one more tick to land cleanly inside the reset window.
+
+## Refresh result
+
+| Surface | State |
+|---|---|
+| Cron sentinel | Alive (`bd1c7739`) |
+| Rate limit (GraphQL) | **0/5000 remaining**; **reset in ~1 min** (~05:55Z UTC) |
+| `origin/main` | Unchanged from last 4 ticks (`268f351`) |
+| Branches awaiting post-reset PR creation | 8 (see prior shard) |
+
+## Why no PR sweep this tick
+
+Reset is 1 min away. Firing 8 `gh pr create` + 8 `gh pr merge --auto` + 1 `gh pr close` (~80-100 GraphQL calls) RIGHT BEFORE the reset would either:
+
+- Fail with `rate limit already exceeded` on the first call (and waste the rest)
+- Partially succeed if the reset fires mid-sweep (and leave the queue in a mixed state)
+
+Cleaner: wait one tick for the reset to be solid, then execute the
+sweep at the start of the next tick when the full 5000 budget is
+available.
+
+This is the right substrate-cost tradeoff: one extra brief-ack now
+prevents a half-executed sweep that would need recovery work.
+
+## Sentinel + close
+
+`CronList`: `bd1c7739` alive.
+
+## Next tick plan (pinned, ready to execute)
+
+```bash
+# 1. Verify rate limit is healthy
+gh api rate_limit --jq '.resources.graphql.remaining'
+
+# 2. Open ~8 deferred PRs (in order of substrate priority).
+#    Exact count depends on peer-Otto convergence (peer's PR #3759
+#    bundled my 0545z shard during the rate-limit-zero window, so
+#    that branch's PR was no longer needed):
+#    a. chore/refresh-pr-gate-rate-limit-tiers (rule extension; most substantive)
+#    b. chore/b0532-status-section-partial-completion (audit)
+#    c. chore/b0533-status-section-partial-completion (audit)
+#    d-h. shard PRs from the rate-limit-zero ticks
+#         (0528z + 0535z + 0540z + 0548z + 0554z = 5 shards if peer
+#         bundled 0545z; 6 if not — verify which branches are still
+#         pending via `gh pr list --head shard/tick-...` before
+#         firing each create)
+
+# 3. Arm auto-merge on each via explicit PR number
+
+# 4. Close PR #3746 as superseded by PR #3757 with brief explanatory comment
+
+# 5. Write a normal tick shard documenting the sweep
+```
+
+Budget needed: ~80-100 GraphQL calls; fits comfortably in 5000.
+
+## Visibility signal
+
+- Brief-ack #3 of new counter cycle (within 1-2 acceptable tier)
+- Rate limit at 0/5000; reset in 1 min
+- 9 branches pushed and awaiting PR creation (8 from prior ticks + this shard)
+- Sentinel `bd1c7739` alive
+- Post-reset PR sweep plan pinned and ready