Sprint 155 — Seed #22 pre-push Hook Staging Auto-Verification (3-Layer Safety Net Completion)

Goals

• Directly reinforce Sprint 153 Phase A/E incidents (PR #241 hotfix) + Sprint 154 §6 specification + Sprint 154 Lesson #4 (untracked verification blind spot)
• Complete 3-layer safety net: plan stage (Sprint 154 A) + pre-push stage (this sprint) + CI lint stage (Sprint 154 B)
• Eliminate untracked .md broken ref blind spot — first detection locally before push

Decisions

• Phase 1 (architect): New .husky/pre-push shim + scripts/check-staging-integrity.mjs. Add --include-untracked option + ESM export to check-doc-refs.mjs
• Phase 3+4 (scribe): New docs/runbook/pre-push-check.md + git-hooks.md §2 expansion + 3 existing RUNBOOK cross-refs + ADR
• Single sprint 2 PR bundle — Phase 1 code (architect) → Phase 3+4 docs (scribe) sequential merge. Sprint 150~154 pattern inherited
• 2 regression fixtures: Sprint 153 Phase E (old runbook slug) / Sprint 154 PR #246 (home path reference) — previous incidents fixed as self-test baseline

Implementation (2 PR squash merge, origin/main 1e51e24 → TBD)

Phase 1+2 Detail — Code Implementation (PR #247)

.husky/pre-push (2 lines):

#!/usr/bin/env sh
npx --no -- node scripts/check-staging-integrity.mjs

scripts/check-staging-integrity.mjs (341 lines):

• Verification 1: git ls-files --others --exclude-standard "*.md" → untracked .md broken refs
  • Reuses check-doc-refs.mjs's validateRef / extractMarkdownLinks / extractBareDocPaths / stripInlineCode
  • Exemption: <!-- staging-check: ignore --> or <!-- doc-ref-lint: ignore --> (both compatible)
  • exit 2 (push blocked when broken refs exist)
• Verification 2: git status --porcelain Y=M lines → detect unstaged modified files
  • [BLOCK] message + exit 1 (push blocked — intentionally unstaged can bypass with git push --no-verify)
• regression fixture runRegressionFixtures():
  • Scenario 1 — Sprint 153 Phase E: old slug docs/runbook-monitoring-log-rules.md reference → 1 violation detected
  • Scenario 2 — Sprint 154 PR #246: home path ~/.claude/projects/.../sprint-999.md → 1 violation detected
  • exit 2 (fail-safe) if self-test fails

scripts/check-doc-refs.mjs expansion:

• Export validateRef / extractMarkdownLinks / extractBareDocPaths / stripInlineCode / collectUntrackedMarkdown
• Entry point guard (process.argv[1] === __selfPath) — main logic does not execute on import
• --include-untracked flag: combined scan of tracked + untracked .md (166 files, 0 broken refs)

Phase 3+4 Detail — Documentation (PR #248, this PR)

New docs/runbook/pre-push-check.md (7 sections):

• §1 Overview — Sprint 153 Phase E / Sprint 154 PR #246 2 incidents directly cited. 3-layer safety net diagram
• §2 Verification items (2 types) — exit code behavior + exemption pattern details
• §3 Exemption policy — automatic exemption patterns + explicit directive (<!-- staging-check: ignore -->)
• §4 Operations procedure — Husky activation / automatic execution on push / manual execution
• §5 Bypass — allowed cases for git push --no-verify explicitly stated
• §6 FAQ — false positive / intentional unstaged / self-test failure response
• §7 History

docs/runbook/git-hooks.md §2 newly created:

• §2.1 Introduction background — Sprint 153 Phase E / Sprint 154 PR #246 incident summary table
• §2.2 Verification item summary + pre-push-check.md cross-ref
• §2.3 Bypass procedure (--no-verify)

Existing RUNBOOK cross-ref updates (3 items):

• git-staging-checklist.md §6: "future expansion candidate" → "Sprint 155 completed" + 3-layer safety net table + ADR cross-ref
• doc-ref-lint.md §8: untracked limitation specified + --include-untracked option + pre-push-check.md cross-ref
• README.md local development environment (4) → (5): pre-push-check item added

Verification

• Phase 1 PR #247 CI: CI SUCCESS, mergeStateStatus CLEAN ✅
• self-test 2/2: Sprint 153 Phase E scenario + Sprint 154 PR #246 scenario manually reproduced → exit 2 each confirmed
• --include-untracked verification: node scripts/check-doc-refs.mjs --include-untracked → 166 files, 0 broken refs
• Local doc-ref-lint: 2 new RUNBOOKs in this sprint (pre-push-check.md, sprint-155.md) passed self-lint (Sprint 154 meta-self-verification pattern inherited)
• Auto-Critic R1 P2 1 issue → R2 clean (Phase 1 architect): [WARN] message + exit(1) inconsistency caught → message/symbol/exit code 3-element consistency policy established

Branch Discipline

• Both PRs use new branches + Squash merge — 22 consecutive sprints compliant (since Sprint 134 violation)
• 0 direct commits to main

New Patterns

1. 3-layer safety net pattern completed (Sprint 154 2-layer → this sprint 3-layer) — plan (checklist) + pre-push (hook) + CI (lint). Each stage blocks a different blind spot at a different point in time. If one stage is bypassed, the next stage defends
2. Untracked supplement bypass mode (--include-untracked) — Added untracked file support to check-doc-refs.mjs's git ls-files based scan. pre-push hook runs automatically; manual verification uses --include-untracked option
3. ESM function export safety via entry point guard — process.argv[1] === __selfPath condition prevents main logic execution on import. Prevents CI entry point duplicate execution when check-staging-integrity.mjs imports from check-doc-refs.mjs
4. Regression fixture directly mapped to preceding incident scenarios — Sprint 153 Phase E (old slug) + Sprint 154 PR #246 (home path) fixed as self-test baseline. Same principle as Sprint 154 pattern (5-slug fixture)

Lessons Learned

1. [WARN] vs exit(1) inconsistency immediately caught by Auto-Critic R1 (preempts user/team confusion) — exit(1) actually blocks push but [WARN] prefix is easily misread as "OK to continue." Message symbol ([OK]/[WARN]/[FAIL]) + exit code 3 elements must be consistent for hook reliability
2. <!-- staging-check: ignore --> and <!-- doc-ref-lint: ignore --> compatibility policy — Both lints inspect the same files, so a single exemption directive must exempt both. Minimizes operational burden
3. Message/symbol/exit code 3-element consistency is the core of hook reliability — [OK] = exit 0 / [WARN] = exit 1 (soft block) / [FAIL] = exit 2 (hard block). Inconsistency risks team members ignoring output or making incorrect judgments
4. New RUNBOOK self-verification via doc-ref-lint (Sprint 154 meta-self-verification pattern inherited) — pre-push-check.md introduced in this sprint also passed lint before commit. The 3-layer safety net RUNBOOK is verified by the net itself

Sprint 156 Carryover

UAT User Direct (12 sprints accumulated)

• Seed #5: Programmers resubmission scoring pass confirmation
• Seed #9: English environment + production Grafana CB dashboard ai-analysis visual consistency

Sprint 152~153 New Automation Candidates (Carry Forward)

• Seed #18: Blog post pre-merge domain fact cross-check automation
• Seed #19: KR/EN dual simultaneous authoring plan obligation + CI rule

Follow-up (Optional, Sprint 151 unchanged)

• create/edit page.tsx category UI addition
• Programmers URL automatic category inference
• Existing SQL problem data backfill
• Sprint 150 unresolved 3 candidates (.claude-tools/ cleanup / CI paths filter bypass debt check automation / prom-client default metric stale check)

Related Memory

• sprint-window.md