Sprint 150 — Carryover Seeds Cleanup (Seeds #16/#15/#14 Bundle Processing)

• Period: 2026-05-13 (single day)
• Status: Completed ✅
• origin/main: d1fe387 → 2f68402 (3 PR squash merges)
• start_commit: d1fe387
• end_commit: 2f68402
• Merged PRs: 3

Goals

Resolve 3 automation candidates from Sprint 149 carryover (seeds #14/#15/#16) in a single sprint. Inherits Sprint 144149's "single-day 23 PR bundle" pattern. Each seed separated into independent PR units for squash merge.

Merged PRs

PR	Seed	Commit	Change Scale	Critic
#226	#16 .claude/commands tracked policy	c08f5e4	21 files / +1161 -4	Not invoked (policy/docs)
#227	#15 extractInlineBlock 6-type YAML modifier	dcd2502	1 file / +29 -2	R1+R2 P2 resolved → R3 clean
#228	#14 problem context coverage strengthening	2f68402	3 files / +228 -1	R1 clean + R2 clean

Work Summary

PR #226 (Seed #16) — .claude/commands Tracked Policy (c08f5e4)

Background: Sprint 147 retrospective revealed limitation that RUNBOOK cross-refs added to .claude/commands/agents/{critic,architect}.md cannot sync to other machines because they're gitignored.

Changes:

• .gitignore: Introduced .claude/* + !.claude/commands/ negation pattern (commands tracked, settings.local.json/scheduled_tasks.lock untracked)
• .claude/commands/ 18 files newly tracked (5 root commands + 13 agents)
• docs/runbook-claude-commands.md new 6 sections — tracked/untracked boundary policy, new registration obligation, security grep checklist, local/shared boundary decision criteria
• CLAUDE.md "Agent Workflow" section expanded
• agents/ stale line removed (.gitignore L92 — directory doesn't exist)
• .claude-tools/ explicitly untracked (Oracle dispatch tools, separate cleanup planned)

Security verification: grep of all 18 files — 14 matches all policy keyword text (0 actual secret values).

PR #227 (Seed #15) — extractInlineBlock 6-Type YAML Modifier Recognition Unification (dcd2502)

Background: Sprint 148 PR #221 Critic R2 observation. scripts/check-grafana-metrics.mjs:638 extractInlineBlock() recognizes only single ${key}: | modifier, while the same file's validateRuleExprLabels (around L972) recognizes all 6 modifiers (|, |-, |+, >, >-, >+) — asymmetric. Currently no regression since all ConfigMaps use |, but future authors using |-/> would cause silent skip → dashboard verification itself would not operate.

Changes (R1→R2→R3 incremental refinement):

• R1 (8741aea): [|>][-+]?\s*$ regex + escapeRegExpLiteral helper — over-narrow (rejects inline comment + indentation indicator)
• R2 (ce12c91): [|>](?:[-+]?[1-9]?|[1-9]?[-+]?)\s*(?:#.*)?$ — allows indentation indicator but conflicts with hardcoded 4-space body indent
• R3 (09ec31b): [|>][-+]?\s*(?:#.*)?$ — intentionally rejects explicit indentation indicator + maintains inline comment. Policy explicitly documented (avoids silent skip)

Verification matrix:

• POSITIVE 9 types: 6 modifiers + 3 inline comment variants → EXTRACT ok
• INTENTIONAL REJECT 5 types: |1, |2, |-2, |2-, |+3 → NULL ok (intentional)
• INVALID YAML 5 types: |*, |abc, |--, ||, |> → NULL ok
• Baseline unchanged: 204 metrics / 32 strict / 15 wildcard / 124 labels / 41 panel pairs / 2 vars / 15 rule pairs / 0 violations

PR #228 (Seed #14) — Problem Context Regression Blocking + Coverage Strengthening (2f68402)

Background: Regression blocking test absence for submission.service.ts L89~90 problemTitle ?? '' fallback policy decided in Sprint 143 Critic R2 P2. Additionally, CI Test Submission was being bypassed by paths filter SKIPPING, exposing debt where functions 95.53% threshold was being circumvented on main.

Change steps:

1. de08975 (submission.service.spec.ts +72): 3 regression blocking tests for ?? '' fallback
   • null title/description → stores '' in entity
   • undefined title/description → stores '' in entity
   • normal title/description → stores actual value in entity
2. f98ff19 (problem-service-client.spec.ts +156 + jest.config.ts +1):
   • getProblemInfo() 5 cases (normal / userId not passed / CB throw fallback / Error resolve defense / config not set immediate fallback)
   • _doGetProblemInfo() 3 cases (200 normal / 200 missing empty string / 404 throw)
   • jest.config.ts: Added !**/*.spec.ts to collectCoverageFrom (policy clarification)

Verification:

• jest 354 tests passed (previous 346 + 8 new), success=true
• All thresholds passed
• problem-service-client.ts coverage: stmt 78.9 → 97.77 / br 72.7 → 93.93 / fns 83.3 → 100 / lines 78 → 98.8

Critic Invocation (3 PRs × Rounds)

PR	Round	Session ID	Result
#226	—	—	Not invoked (policy/docs)
#227	R1	019e1ebb-8a05-7543-ba4f-0ccfee5cb1cc	P2 1 (anchored regex over-narrow)
#227	R2	019e1ebe-45a2-7161-a603-9e303156fbec	P2 1 (indentation indicator + body indent)
#227	R3	019e1ec0-f336-74e3-926b-67d8559b7f4f	clean ✅
#228	R1	—	clean (focused regression tests)
#228	R2	—	clean (coverage strengthening + spec exclusion policy)

Verification Results

PR	CI	Test Submission	mergeStateStatus
#226	38 SUCCESS / 11 SKIPPED	n/a (paths SKIPPED)	CLEAN
#227	27 SUCCESS / 12 SKIPPED	n/a (paths SKIPPED)	CLEAN
#228	28 SUCCESS / 11 SKIPPED	pass (354 tests, threshold passed)	CLEAN

Branch Discipline

• All 3 PRs: new branches + Squash merge — 16 consecutive sprints compliant (since Sprint 134 violation)
• 0 direct commits to main
• Branch names: chore/sprint-150-claude-tracked, refactor/sprint-150-extract-inline-block-symmetric, test/sprint-150-problem-context-coverage

New Patterns

1. .claude/commands/ Tracked SSOT Conversion

12 agent personas + 5 root command files converted to tracked via .gitignore negation pattern (!.claude/commands/). Sync possible across multiple machines/teammates. Local-only files (settings.local.json, scheduled_tasks.lock) remain untracked with .claude/*, making the boundary explicit.

2. Regression Blocking Core Cumulative Dimension Extension — 7th Dimension

Following Sprint 145~149 monitoring verification accumulation (metric → label → panel-title+variable → rule-label+dashboard-structure → regex-robustness), submission service problem context fallback SSOT verification closed. 4-layer cumulative verification:

• submission.entity (nullable)
• submission.service.ts L89~90 (?? '' fallback)
• worker.py L220~227 (null branch, Sprint 143)
• submission.service.spec.ts (regression blocking test, this PR)

3. Paths Filter Bypass Debt Exposure Mechanism

When CI Test Submission is SKIPPED by paths filter (services/submission/**), main's coverage threshold failures are bypassed. Actual execution occurs in PRs where submission/** changes are made, exposing failures → immediate strengthening obligation. This time problem-service-client.ts's getProblemInfo function itself was not called in unit tests.

4. Critic 3-Round P2 Resolution → R3 Clean Pattern (PR #227)

• R1 over-narrow → R2 over-permissive (allows indentation indicator → body indent assumption breaks) → R3 accurate (intentional reject)
• R3 conclusion: "Explicit indentation indicator (|1/|2) is unused in production ConfigMaps → explicit reject avoids silent skip"
• Same pattern as Sprint 142 (5 rounds) / Sprint 148 (3 rounds) / Sprint 149 (4 rounds) — single condition incremental refinement

5. Seed Bundle Single Sprint Processing Effect

3 carryover seeds processed in a single day with 3 PRs — inherits Sprint 144 pattern (2 PRs / 21 minutes). Fresh context + co-exposure of incidental debt (coverage threshold) + immediate retrospective seed processing.

Lessons

1. Paths Filter is a Double-Edged Sword — Accumulated Debt Bypass Risk

detect-changes paths filter is effective for reducing CI time, but services with no changes can bypass coverage threshold failures. Metrics passing on main may actually fail when executed. → Regular check (e.g., quarterly full CI run) or separate weekly job candidate identified.

2. Adding Specs Alone Cannot Strengthen Coverage

Adding 3 regression blocking tests to submission.service.spec.ts alone does not cover problem-service-client.ts production functions themselves. The core of regression blocking is executing production branches, so unit tests for production methods are needed beyond mock factory additions.

3. Single-Function Modifier Support Range Asymmetry is a Time Bomb

extractInlineBlock single modifier vs validateRuleExprLabels 6-modifier asymmetry was recognized in Sprint 148 but classified as gradual improvement candidate. Resolved in this PR 6 sprints later. Silent skip risk accumulates from the recognition point.

4. Defect Patterns Frequently Caught by Critic Are CI Automation Candidates

PR #227 R1+R2 P2 catches both follow patterns explicitly stated in RUNBOOK §2.4 prefix anchoring checklist. Sprint 149 PR #224 completed lint automation (check-regex-robustness.mjs). Future same patterns blocked at lint stage before Critic invocation.

5. Operational Policy Documentation is Key to Avoiding Silent Skip

YAML 1.2 spec allows |1/|2 indentation indicators as valid, but they're unused in this codebase. R2 fix allowing them conflicted with hardcoded body indent. R3 policy documentation (unsupported) + explicit NULL return when someone uses them in the future → intentional safety net.

Sprint 151 Carryover Seeds

UAT User Direct (Outside Oracle's Scope)

• Seed #5: Programmers resubmission scoring pass confirmation (7 sprints accumulated)
• Seed #9: English environment + production Grafana CB dashboard ai-analysis visual consistency

Automation / Infrastructure

• No new carryover from this sprint: 0 (all 3 Sprint 149 automation candidates resolved)

Candidates (Separate Decision Needed)

• .claude-tools/ Oracle dispatch tool cleanup (explicitly untracked in PR #226, separate seed)
• CI paths filter bypass debt check automation (lesson #1 from this PR)
• prom-client default metric stale check (Sprint 145 seed #10)