Sprint 128 — Sprint 127 Follow-up + Oracle SSOT Integration Wrap-up

Background

Closing 4 seeds from Sprint 127 (admin/feedbacks concurrency P2 2 items, Oracle Wave F flag not connected, AnalyticsCharts type narrowing) in a single session. All are follow-up enhancements, not regression fixes, and are items handed over from Critic 2nd/3rd observation.

PM principle: Following the PR #153 close decision in Sprint 127 Wave F, wrapping up the remaining "policy + enforcement" SSOT unintegrated state in this sprint. Connecting the .claude-team.json flag and oracle-reap.sh validation logic via env var so both point to the same truth. Unlike PR #153 which only added the flag, enforcement (reap logic) now works together.

Sprint 128 Processing Status

Wave A — admin/feedbacks SWR Concurrency Remaining P2 2 Items (commit ed5d831)

Assigned: architect (implementation), critic (planned — cross-review before merge)

Background (Sprint 127 Critic 2nd Observation)

In Sprint 127 Wave B, the revalidate: false pattern blocked GET preceding race for single PATCH, but two remaining scenarios were identified in Critic 2nd review:

1. Filter consistency (P2): With statusFilter='OPEN' active, changing a row to RESOLVED → row persists in optimistic stage → suddenly disappears after PATCH completion triggers GET revalidation (unnatural UX). counts card also not synchronized.
2. Concurrent PATCH race (P2): Quickly changing fb-1/fb-2 two rows → PATCH2 response arrives before PATCH1 → GET triggered by PATCH2 success reflects PATCH1 incomplete state (server still has fb-1 as OPEN) in cache → cache inconsistency.

A-1 — listUpdater Filter Consistency

Added statusFilter closure capture to listUpdater inside handleStatusChange. Added filter step after items.map (exclude rows where statusFilter !== 'ALL' && fb.publicId === publicId && newStatus !== statusFilter). Extracted previous status with current.items.find(...)?.status and performed optimistic counts -1/+1 update (category counts like cat:BUG are unaffected by status changes, so they are preserved).

A-2 — useRef in-flight Counter

Option comparison:

Adopted: inFlightRef. +1 on PATCH start, -1 in finally block. Only call mutateFeedbacks() + mutateDetail() when 0 → GET revalidation triggered only after all PATCHes complete.

Additional simplification: removed existing isDetailOpen branching closure. Always call mutateDetail() when in-flight is 0 (safe since SWR is no-op when selectedPublicId is null). Avoids potential omission cases when two PATCHes have different isDetailOpen values.

A-3 — 5 New Regression Tests

Reused existing deferred promise + mockFetcher.mock.calls.length counting pattern.

• Filter consistency 3 items: row removal on out-of-scope change / row retention with ALL filter / counts -1+1 update
• Concurrent race 2 items: PATCH2 resolves first → GET pending before PATCH1 completes / PATCH failure finally counter decrement

Total admin/feedbacks tests: 22 → 27 passing, Jest global: 1388 → 1398 passing.

Change Summary

• 2 files, +196/-14
  • frontend/src/app/[locale]/admin/feedbacks/page.tsx (handleStatusChange restructuring, +28/-14)
  • frontend/src/app/[locale]/admin/feedbacks/__tests__/page.test.tsx (+168 new describe blocks x2)

Wave B — Oracle SSOT Integration (commit d4c27a1 + local infra)

Assigned: architect (.claude-team.json + oracle-spawn.sh + oracle-reap.sh), scribe (critic.md wording sync)

Background (Sprint 127 Wave F PR #153 Close Reason)

Sprint 127 Wave F introduced hardcoded UUID validation in oracle-reap.sh and separately added .claude-team.json dispatch.autoCritic.requireSessionId: true flag, but flag and validation logic were not connected — Critic 1st (019dc216) pointed out "no enforcement" → PR #153 closed. SSOT maintained as oracle-reap.sh alone.

Sprint 128 goal: Reintroduce flag + connect reap logic branching for "policy + enforcement" SSOT integration on both sides.

B-1 — .claude-team.json Flag Reintroduction

"autoCritic": {
  "enabled": true,
  "trigger": "commitDetected",
  "method": "codex review --base <HEAD_BEFORE>",
  "requireSessionId": true
}

Default true preserves existing behavior. Opt-out requires explicit false setting.

B-2 — oracle-spawn.sh env Passing + oracle-reap.sh Flag Branching

oracle-spawn.sh cleanup trap (just before calling oracle-reap.sh):

export REQUIRE_CODEX_SESSION_ID=$(jq -r '.dispatch.autoCritic.requireSessionId // "true"' \
  "${project_dir}/.claude-team.json" 2>/dev/null || echo "true")

oracle-reap.sh UUID validation block (L89~95):

local require_sid="${REQUIRE_CODEX_SESSION_ID:-true}"
if ! grep -qiE 'session id: <UUID>' "$inbox_file" 2>/dev/null; then
  if [[ "$require_sid" == "true" ]]; then
    status="failed_no_codex_session"
    warn "...→ Refusing Claude-only analysis"
  else
    warn "...(requireSessionId=false → warning only)"
  fi
fi

dry-run validation results:

• REQUIRE_CODEX_SESSION_ID=true + UUID missing → failed_no_codex_session ✅
• REQUIRE_CODEX_SESSION_ID=false + UUID missing → completed (warning only) ✅
• Default (no env) → treated as true preserving existing behavior ✅

B-3 — critic.md Wording Sync

Toned down L50 **Session ID Validation** entry and L70 report format "required, result invalid if absent" to "Required by default (toggleable via .claude-team.json requireSessionId flag). Flag changes require Architect consultation". Actual enforcement delegated to flag and reap logic.

Change Summary

• git tracked: .claude-team.json 1 file, +2/-1
• Local infra (gitignored): oracle-spawn.sh env export 1 line + oracle-reap.sh branching 6 lines + critic.md wording 2 lines

Wave C — AnalyticsCharts Type Narrowing (commit e92fad8)

Assigned: herald (type + test fixture)

Background (Sprint 127 Critic 3rd Observation)

In Sprint 127 Wave A, the Difficulty ↔ DifficultyDisplay enum separation was completed, but AnalyticsCharts.tsx's DifficultyRow.tier remained as string with unsafe cast (DIFFICULTY_LABELS[row.tier as DifficultyDisplay] ?? row.tier) remaining. Handed over as Sprint 128 seed.

C-1/C-2 — Type Narrowing + Cast Removal

- interface DifficultyRow { tier: string; count: number; color: string; }
+ interface DifficultyRow { tier: DifficultyDisplay; count: number; color: string; }

- DIFFICULTY_LABELS[row.tier as DifficultyDisplay] ?? row.tier
+ DIFFICULTY_LABELS[row.tier]

DifficultyDisplay is already imported, so no new import needed. Since DIFFICULTY_LABELS defines all 7 entries (including UNCLASSIFIED), the ?? row.tier fallback can also be safely removed.

C-3 — Test Fixture UPPER_CASE Alignment

Analytics.test.tsx L56~59 fixture was written with 'Silver'/'Gold' (PascalCase), inconsistent with the new type. Changed to 'SILVER'/'GOLD' (UPPER_CASE) + 1 'UNCLASSIFIED' case added. Total solution count assertion 14 → 17.

C-4 — Data Supplier Verification

analytics/page.tsx L209~221 useMemo difficultyData already supplies DifficultyDisplay type values as tier: d.key — no changes needed. Verified tsc --noEmit passes after type narrowing.

Change Summary

• 2 files, +6/-5 (AnalyticsCharts.tsx type + branching simplification, Analytics.test.tsx fixture)
• analytics Jest: 6 tests passing

Critic Cross-Review Cycle (codex)

Total 3 cycles: P2 2 items + P3 1 item found → immediately fixed → 3rd cycle passed:

Verification Results

• npx tsc --noEmit: passed
• npx next lint: 0 new warnings (only existing inline style warnings remain)
• npx jest: 130 suites, 1400 tests passing (1388 → +12 new: admin/feedbacks 7 + analytics 5)
• Wave B dry-run: flag true/false both scenarios behaved as expected
• Critic 3rd codex passed (UUID forced validation passed)

Agent Collaboration

Follow-up Seeds (Sprint 129 — new follow-up, not re-carried)

• (none) All 4 seeds closed in Sprint 128. Can add more if Critic cross-review reveals new findings.

Learning Notes

• Local infra + git tracked SSOT separation: .claude/ is gitignored, ~/.claude/ is external environment. Connecting flag (repo) and enforcement logic (local script) via env var allows maintaining single truth even across separate locations. This pattern can be reused for other policy-enforcement separation cases.
• Closure-captured isDetailOpen trap: Branching dependent on closure variables in async finally blocks risks stale snapshots during concurrent calls. Simpler to use the in-flight counter to decide synchronization timing only, and delegate branching to SWR no-op.