…VISION + IDENTITY sync

Sprint 166 sonrası user-facing doc sync (5 paralel doc-writer agent dispatch,
Sprint 164 patterni). Ground-truth verified: 15 agents (Bug Y2 anchor),
21 skills, 27 MCP tools, 55+ CLI, 46 ADRs, 89.33% coverage, ~16,434 tests.

- docs/ROADMAP-GOD-LEVEL.md (+78/-4): Sprint 165 + Sprint 166 + Sprint 167+168
  timeline. Last update + Next audit refreshed. Sprint 167 P0 4 bug forensic
  (Bug E spawn-lock, Bug G OOM, Bug Z2 Files parser, Bug Z3 memory rebuild).
- BETA-TRACKER.md (+136 net) + BETA-TRACKER-TR.md (+104 net): Gate table
  17/20 PASS+1 FAIL → 19/20 PASS+1 PENDING. Gate #2 vitest closure via
  Sprint 165 T3. Gate #16 ADRs 46. Sprint 165 + 166 detailed summaries +
  Sprint 167+168 timeline.
- DECKENT-MASTER-BLUEPRINT.md (+110/-30): ADR-046 Step Ordering Contract
  documented (§5.1.1), Data Integrity Closure (§5.1.2 — Bug U/V/M), Living
  Docs Pipeline (§5.1.3 — Bug P/Q/W). Version 3.0 → 3.1. Sprint history
  extended.
- README.md + README-TR.md (~16 lines each): badges (tests 12485→16434,
  sprints 164→166), "What's New in Sprint 166" mini section.
- VISION.md + VISION-TR.md (+5/-5 each): **Bug Y2 critical fix** — L19
  "16 agents" → "15 agents" + extended snapshot (27 MCP tools, 46 ADRs,
  Memory V2). Sprint 165 + 166 milestone paragraphs. Phase 2 → Sprint 83-166.
- .deckent/workspace/IDENTITY.md (+13/-9): Sprint metric → sprint-166, MCP
  22→27, CLI 49+→55+, ADRs 46, Sprint 165/166 features appended.

Verify: grep -c "16 agent|test-writer" current-state assertions → 0.
Historical sprint references (Sprint 148 reform, Sprint 73/146) preserved
as factual records.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…on — 17 madde fix

Sprint 168 v1→v4 zinciri:
- v1 (commit fc91fcd): brainstorming output
- v2 systematic-debugging eval (Agent A): 79/100, Phase 4.5 trigger, 5 critical/high
- v3 devil's advocate eval (Agent B): 22/100 — hedef <30 KARŞILANDI ✅
- v4 (bu commit): 17 madde fix integration
- v5 Alperen final approval bekliyor

17 madde fix entegre:
1. (Agent A #1 CRITICAL) C0e cross-sprint orphan handling — Option C selective filter + startup invocation
2. (Agent A #2 CRITICAL) C0a bundle split → C0a-1/C0a-2/C0a-3/C0a-4 (4 sub-anchor)
3. (Agent A #3 CRITICAL) C0c subscriber owner: decision-engine.ts designate + function signature + BRAIN→SPAWN:BLOCKED event mandatory test
4. (Agent A #4 + B V7 HIGH) ADR-047 Manuel Subagent Dispatch Protocol Sprint 168'de (önceki 168.5 ertelenmiş)
5. (Agent A #5 + B Saldırı #2 HIGH) Smoke test complex scenario: 3+ task (collision + crash + parallel)
6. (B V5 HIGH) Sprint 168 NO_GO → Sprint 168.5 fallback explicit (recursion paradox kabul)
7. (B V7 HIGH) TDD skip enforcement gate (skip artış 0 + Alperen review)
8. (Agent A #8 MED) checkSpawnLock singular helper eklendi
9. (Agent A #9 MED) auto_archive_directives Alperen decision NOW (spec yazımı sırasında)
10. (Agent B #5 MED) Subagent git branch isolation (worktree per cluster)
11. (Agent B #7 MED) ADR-046 invariant test C0a-2 + C0a-3 integration test
12. (Agent B #6 MED) ADR-048 multi-provider parity (Docker + Subprocess + Tmux)
13. (Agent B #4 MED) ADR-048 Wave 1.5 serial gate + Alperen CHECKPOINT
14. (P4.5 MED) Cross-cluster dependency graph spec içinde explicit (Section 2)
15. (Agent B #3 MED) Baseline tolerance "0 yeni skip" GO/NO_GO row
16. (Agent B V6 LOW) Effort yeniden tahmin 22-30h → 35h gerçekçi
17. (Agent B #1 LOW) Brain "kırık" iddiası modül-fonksiyon-satır kanıtı (Section 1)

v1 → v4 büyük yapı değişiklikleri:
- Scope 5 task → 8 task (C0a split + ADR-047 yeni)
- Effort tahmin 22-30h → 35h
- ADR sayısı 1 (ADR-048) → 2 (ADR-047 + ADR-048)
- Smoke test 2-task echo → 3+ task complex (collision + crash + parallel)
- Subagent dispatch "manuel" → "hardened (worktree + file authority + lock + TDD gate)"
- Sprint 168.5 dependency açıklama (NO_GO → manuel dispatch replay)
- Section 2 cross-cluster dependency graph yeni
- Section 3.2 (dispatch mechanism + lock pattern + TDD gate + fallback) yeni
- Section 5.3 complex smoke test suite yeni
- Pre-Flight checklist 11 → 14 madde (git worktree, dispatch locks, smoke test, Alperen auto_archive decision)

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

… (çift hedef başarılı)

v4 üzerine 2nd round eval patch — Sprint 167 paterni gibi (v1→v5):
- v4 (72b4880): 17 madde 1st round integration
- v5 2nd round Agent A: 96/100 APPROVED ✅ (hedef ≥95)
- v5 2nd round Agent B: 26/100 SHIP_AS_IS ✅ (hedef <30 korundu)
- v5 patch (bu): 6 madde minor — 4 Agent A cosmetic + 2 Agent B clarification

6 madde patch:
1. (Agent A 2nd round #2) Section 1 — 10 bug attribution explicit per cluster:
   Cluster A 4 bug + B 1 + C 3 + D 1 + E 1 = 10 ✓
2. (Agent A 2nd round #1 CRITICAL miss) C0e scope — getActiveWorkerIds() public
   helper extract instruction (auditor.ts:2162-2168 → src/core/active-workers.ts)
3. (Agent A 2nd round MINOR-3) Section 5.1 prompt template reword —
   "Skip artış YASAK" → "Yeni test'lerde skip kullanma" daha net
4. (Agent A 2nd round Section 3.4 önerisi) GO_WTD ≤1 candidate netleştir —
   en muhtemel C0d cosmetic metrics fix
5. (Agent B 2nd round Saldırı #2) Section 5.3 — Sprint 168.5 pre-flight smoke test
   eklendi (multi-file + DB write + repo scan pattern coverage)
6. (Agent B 2nd round Saldırı #3) Section 8 — Sprint 168.5 wave split opsiyonu
   (168.5a + 168.5b auto-split threshold ≥10 task)
+ (Agent A 2nd round MINOR-4) Pre-Flight checkbox count netleştir: 16 madde

Sprint 168 eval final:
- Agent A 1st 79 → 2nd 96 (+17 puan, APPROVED ≥95 ✓)
- Agent B 1st 22 → 2nd 26 (+4 puan, SHIP_AS_IS <30 korundu ✓)
- v5 patch sonrası spec çift hedef başarılı + minor refinement

v6 Alperen final approval bekliyor — sonra writing-plans skill Sprint 168 TDD plan.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…ical docs

Sub-project #1/4 (Embedded Web Terminal) shipped 2026-05-19→2026-05-20,
operationally smoke-confirmed by Alperen. Update reader-facing docs to
reflect the delivery honestly + the four-part path ahead.

- README + README-TR: new Highlights bullet (top), link to terminal guide
- VISION + VISION-TR: extend 'Where we are now' / 'Where we are going'
  with the embedded terminal as the first concrete step toward agentic-
  OS workflows + the 3 deferred sub-projects (#2 self-security,
  #3 multi-tenant/k8s, #4 enterprise integrations) with the seams
  (AuthProvider / SessionBackend / tenantId) called out
- docs/release/roadmap.md: new 'Phase 3.6: Embedded Web Terminal' before
  Phase 3.5, with delivered checklist + sub-project #2-4 backlog
- docs/release/beta-tracker.md + -tr.md: Last-updated header refreshed
  (2026-05-14→2026-05-20); new 'Sprint 175 — Embedded Web Terminal'
  section with metrics (46/46 tests, build/pack clean), honest debt
  (node-pty linux-x64 prebuild, DECKENT_API_AUTH_DISABLED=1 dashboard
  precondition), process learnings (two durable feedback memories)
- docs/ROADMAP-GOD-LEVEL.md: new ⚡ 2026-05-20 section at the top, full
  Sprint 175 record + sub-project backlog table + Sprint 176+ priorities
- docs/adr/README.md: regenerated index (already had ADR-062/063 after
  collision rename earlier)

No exaggeration: smoke confirmed, but DECKENT_API_AUTH_DISABLED=1 is
still required for dashboard data calls (frontend general API auth
plumbing is sub-project #2/#3 scope) and node-pty linux-x64 prebuild
needs the optionalDep fix in Sprint 176. Recorded as honest debt.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…ems)

The §1d backlog had 3 items captured during Sprint 175 prep; the
post-implementation review surfaced 3 more (4–6) plus a self-security
procedure scope section for the #2 spec phase. All references checked
against current source (npm run lint:link clean):

4. Schema-gate coverage enforcement gap — coverage_threshold auto-lowers
   when avg <70%, so the gate moves with reality instead of holding the
   bar. Needs hard-floor vs aspirational split.
5. WorkerCard/DashboardPage pre-existing TS errors — TS2345 i18n
   contravariance, suppressed because root `lint` doesn't recurse into
   src/dashboard/tsconfig.json. Wire the dashboard tsc into root lint.
6. `doctor` DECISIONS.md obsolete check — Memory V2 (Sprint 143) moved
   this to memory.db; doctor still hard-requires .brain/DECISIONS.md.
   Cascade fossils across constants.ts, debt-manager.ts, sprint-docs-
   helpers.ts, authority-enforcer.ts catalogued.

Self-security scope section (sub-project #2): prompt guard, command
guard, outbound rate-limit, mutual-TLS hook on AuthProvider, self-audit-
of-audit. To be ADR'd in the #2 spec phase.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

systematic-debugging Phase 1+4.5: 5-sprint loop root cause = auto-debt-
injection (sprint-planner.ts:197-216) producing empty-scope CRITICAL
tasks; workers cannot resolve a bookkeeping artifact; re-injection 4
sprints (170→174) was the 3+ failed identical-mode fixes pattern.

Original 170-001 code verified in repo (tmux.ts:61-70, commit 5ffbf3e).
Structural cause already fixed (Sprint 138 T-13 Docker HB Core Fix).

Non-destructive DB upsert: status active→resolved, metadata.resolution
attached (1020 char honest closure). Unblocks Sprint 175 dry-run:
21→20 tasks, dependency IDs realigned (W0.1=175-001, off-by-one fixed).

Architectural follow-up (empty-scope auto-inject) deferred to sub-project
#2 per Alperen 2026-05-20.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…cts)

Caught during Sprint 175 prep, deferred to #2 per Alperen 2026-05-20:
(1) auto-debt-injection empty-scope bug — sprint-planner.ts:197-216,
4-sprint loop root cause for debt-170-001-fix; (2) re-plan orphan
cleanup — .tasks/task-{id}-*.json from prior iteration not unlinked
(today's finding: orphan task-175-021.json hand-removed).

.tasks/ is gitignored — orphan removal not in git history; spec §1d
backlog is the durable record.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Observed during Sprint 175 EXECUTE (user reported terminal warning).
Three call-sites pass args array WITH shell:true on all platforms:
- src/core/plugin-hooks.ts:395, :577
- src/orchestra/baseline-tracker.ts:85
Node DEP0190 + ADR-006 violation + already flagged by
authority-enforcer.ts:464-481 (lint, not runtime-enforced).
Deferred to sub-project #2 per Alperen 2026-05-20.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…ical docs

Sub-project #1/4 (Embedded Web Terminal) shipped 2026-05-19→2026-05-20,
operationally smoke-confirmed by Alperen. Update reader-facing docs to
reflect the delivery honestly + the four-part path ahead.

- README + README-TR: new Highlights bullet (top), link to terminal guide
- VISION + VISION-TR: extend 'Where we are now' / 'Where we are going'
  with the embedded terminal as the first concrete step toward agentic-
  OS workflows + the 3 deferred sub-projects (#2 self-security,
  #3 multi-tenant/k8s, #4 enterprise integrations) with the seams
  (AuthProvider / SessionBackend / tenantId) called out
- docs/release/roadmap.md: new 'Phase 3.6: Embedded Web Terminal' before
  Phase 3.5, with delivered checklist + sub-project #2-4 backlog
- docs/release/beta-tracker.md + -tr.md: Last-updated header refreshed
  (2026-05-14→2026-05-20); new 'Sprint 175 — Embedded Web Terminal'
  section with metrics (46/46 tests, build/pack clean), honest debt
  (node-pty linux-x64 prebuild, DECKENT_API_AUTH_DISABLED=1 dashboard
  precondition), process learnings (two durable feedback memories)
- docs/ROADMAP-GOD-LEVEL.md: new ⚡ 2026-05-20 section at the top, full
  Sprint 175 record + sub-project backlog table + Sprint 176+ priorities
- docs/adr/README.md: regenerated index (already had ADR-062/063 after
  collision rename earlier)

No exaggeration: smoke confirmed, but DECKENT_API_AUTH_DISABLED=1 is
still required for dashboard data calls (frontend general API auth
plumbing is sub-project #2/#3 scope) and node-pty linux-x64 prebuild
needs the optionalDep fix in Sprint 176. Recorded as honest debt.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…ems)

The §1d backlog had 3 items captured during Sprint 175 prep; the
post-implementation review surfaced 3 more (4–6) plus a self-security
procedure scope section for the #2 spec phase. All references checked
against current source (npm run lint:link clean):

4. Schema-gate coverage enforcement gap — coverage_threshold auto-lowers
   when avg <70%, so the gate moves with reality instead of holding the
   bar. Needs hard-floor vs aspirational split.
5. WorkerCard/DashboardPage pre-existing TS errors — TS2345 i18n
   contravariance, suppressed because root `lint` doesn't recurse into
   src/dashboard/tsconfig.json. Wire the dashboard tsc into root lint.
6. `doctor` DECISIONS.md obsolete check — Memory V2 (Sprint 143) moved
   this to memory.db; doctor still hard-requires .brain/DECISIONS.md.
   Cascade fossils across constants.ts, debt-manager.ts, sprint-docs-
   helpers.ts, authority-enforcer.ts catalogued.

Self-security scope section (sub-project #2): prompt guard, command
guard, outbound rate-limit, mutual-TLS hook on AuthProvider, self-audit-
of-audit. To be ADR'd in the #2 spec phase.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

PR #16 main merge surfaced 3 test failures that all pass locally on the
same Node 24.15.0 runtime — pure CI environment divergence. Captured in
sub-project #2 backlog as item #7 (lokal-CI gap) with concrete fix
candidates (mock hygiene, process.kill semantics, vitest --retry tag).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…rity

Sprint 176 spec — 12 task, 5 wave, June 1 2026 beta-blocker:

A. Planner state-hygiene (W1-W3, 7 task):
   1. Auto-debt empty-scope (sprint-planner.ts:197-216)
   2. Re-plan orphan task file cleanup
   3. DEP0190 shell:true (3 call-site)
   4. Coverage hard-floor / aspirational split
   5. Dashboard TS errors + root lint wire
   6. doctor DECISIONS.md obsolete + 5-file cascade
   7. CI-only test flakes (lokal-CI divergence)

B. Self-security guards (W4-W5, 5 task — interceptor pattern, 5 new files):
   8. Prompt guard (input pattern matcher)
   9. Command guard (shell-kind deny-list, remote-only)
   10. Outbound rate-limit (daily tenant-scoped quota)
   11. mTLS hook (AuthProvider interface extension)
   12. Self-audit-of-audit (HMAC append-only chain + verify CLI)

Five non-negotiable security invariants (I1-I5):
- I1 no silent drop, I2 no raw payload in audit, I3 default-deny
  on host≠127.0.0.1, I4 append-only audit + HMAC tamper detect,
  I5 tenant-scope isolation (audit + quota + guard state).

Locked decisions: single sprint #176, structured planning, sequential
self-modifying dispatch (Sprint 175 pattern), Brain manuel wave gates
(ADR-047), non-destructive memory.db ALTER, GO_WITH_TECH_DEBT acceptable
verdict (≤2 GWT, W1-W3≥5 DONE, W4≥2 DONE).

Predecessor: PR #16 (sub-project #1, terminal, merged 2026-05-20).
Successor: sub-project #3 (multi-tenant + k8s + mTLS impl, post-beta).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…down

759-line plan covering 12 task across 5 sequential waves (W1-W5):
W1 planner P0 (auto-debt + orphan cleanup), W2 discipline gate (DEP0190
+ coverage split + CI flakes), W3 Memory-V2 cascade + frontend
(dashboard TS + doctor cascade), W4 self-security core (prompt-guard,
command-guard, outbound-limiter — I1-I3+I5), W5 self-security ileri
(mTLS hook + audit HMAC chain — I4).

Each task carries bite-sized RED->GREEN->REFACTOR steps, exact file
paths + line numbers from spec §1d, GO/NO_GO criteria, and commit
message templates. Plan also includes ready-to-paste DIRECTIVES.md
content for Sprint 176 launch via deckent_set_directives.

Refs spec docs/superpowers/specs/2026-05-21-sub-project-2-design.md.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…rift

Sprint 176 worker on task 176-001 produced a 3 KB NO_GO refuse note
documenting auto-debt-injection empty-scope failure — a live reproduction
of spec §1d.1, the exact bug W1-1 fixes. On SIGTERM the brain-honest-
gate mechanism stub-overwrote the result file to 364 bytes erasing the
forensic notes.

This document restores the original notes from the conversation
transcript + records the two concurrent dogfood bugs that triggered the
kill: (1) config.json template-regen losing spawn_backend (cascade from
PR #16 git rm --cached); (2) nervous_system.directives_protection
auto_restore rolling DIRECTIVES.md back to Sprint 175 after cleanup.

Cross-referenced in sub-project #2 plan Task 1 — to be cited in the
regression fixture for sprint-planner debt injection. Open question
captured for sub-project #2 backlog #8 (nervous_system baseline-update
hook on deckent_set_directives success).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Sprint 176 dogfood produced 8 simultaneous failure modes (documented in
docs/audits/sprint-176/dogfood-evidence.md). The original sub-project #2
spec only addressed item 7 (planner state-hygiene + self-security); this
master spec covers all 8 + remaining feature backlog, decomposed into 4
sprints with beta-gate prioritization.

Locked decisions (Alperen 2026-05-21):
- Worker rollback = git stash --include-untracked snapshot-on-spawn;
  NO_GO drops stash + reverts worker scope writes.
- Tmux backend = deprecate. Sprint 177 marks @deprecated + warning;
  Sprint 178 removes code path.

Sprint 177 = 5 tasks (critical runtime stability):
  177-001 Worker rollback (git stash snapshot-on-spawn)
  177-002 deckent kill cascade fix (worker → controller → metadata)
  177-003 Tmux backend deprecate path
  177-004 Config template-regen guard + restore docs
  177-005 nervous_system directives_protection baseline-update hook

Sprint 178 = 4 tasks (Node 24/26 spread + tmux code removal + CI flakes).
Sprint 179 = 12 tasks (original sub-project #2 scope, references
2026-05-21-sub-project-2-design.md unchanged).
Sprint 180 = nervous defaults + dashboard + feature backlog (post-beta).

Beta gate (June 1 2026) MUST: items 1, 2, 3, 4a-b, 5 (Sprint 177) +
item 7c self-security (Sprint 179 W4-W5). Total beta-required: 10 tasks
in ~7 days, with worker rollback live from Task 177-001 onward.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…m 6e)

Crisis Stabilization master spec missed the TOPP continuous-dispatch
work documented in memory project_topp_continuous_dispatch.md (Alperen
onayı 2026-05-19). Without TOPP, Sprint 179's 12-task sub-project #2
scope runs wave-barrier-throttled (~2-3 effective parallel slots even
with max_workers=6) putting the June 1 beta gate at risk.

Changes:
- §2 inventory: add item 6e (TOPP) — beta MUST
- §4 Sprint 178 scope: 4 task → 5 task (added Task 178-005 with full
  implementation details: result-collector.ts:380 dispatchTick flag-
  agnostic, sprint-spawner.ts:472/509/296-313 continuous body + initial
  fill, prompt-god-template.ts:291-307 TOPP C predecessor digest embed,
  DECKENT_LEGACY_FIFO=1 escape hatch, new ADR contract-first)
- §7 beta gate analysis: TOPP factored into Sprint 178 timing + Sprint
  179 day estimate (3-4 with TOPP vs 5-6 without)

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Sprint 177 CLEANUP/RETRO phase auto-generated bookkeeping:
- docs/CHANGELOG.md + docs/SPRINT-LOG.md + docs/vision/roadmap.md +
  docs/ROADMAP-GOD-LEVEL.md: Sprint 177 entry appended by sprint-docs-
  updater.
- .deckent/ci-baseline.json + features-manifest.json + provider-cache.json:
  runtime stats refresh.
- .deckent/archive/metrics/metrics-sprint-{165,167}.jsonl.gz: aged out
  per sprint_file_retention.keep_last_n=10 policy.
- .gitignore: add .tmp/ (Sprint 177 worker runtime artifact pattern;
  sub-project #2 backlog cleanup-debris #8 follow-up).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Sprint 178 delivers Crisis Stabilization §4 in 35m 11s with 9 DONE +
0 debt + 2 NO_GO (auto-debt 175-020 cluster only). +965 / -131 across
src + tests + docs + 1 new ADR. Coverage 95% on TOPP impl.

Highlights:
  178-007 ★ TOPP B+C continuous-dispatch — DONE on FIX retry
    src/orchestra/result-collector.ts: NEW planDispatch(state) pure
    function (flag-agnostic, returns DispatchPlan { toSpawn, toKill,
    mode }) supersedes ADR-045 §3 wave-barrier semantics. dispatchTick
    closure-scoped async wrapper replaces dual await processQueue +
    maybeRespawn sequence. 440-line G1-G10 test matrix.
    docs/adr/064-topp-continuous-dispatch.md (NEW): ADR contract first.

  178-003 Node 24/26 test assertion sweep — DONE
    4 test files: publish-workflow.test.ts, publish.test.ts, install-
    matrix/fresh-install.test.ts ('22.x' → '24.x', NODE_VERSIONS
    [18,20,22] → [24,26], engines.node >= 24), release-prep.test.ts.

  178-004 Doc updates Node 24/26 yayılma — DONE
    docs/guide/{deckent-nedir,getting-started,quickstart}.md +
    docs/CHANGELOG.md + docs/SPRINT-LOG.md engines refs.

  178-005 Tmux removal — DONE on FIX retry
    src/core/multi-ide.ts + orphan-cleaner.ts + src/orchestra/
    heartbeat-daemon.ts + sprint-phases.ts + sprint-pid-manager.ts
    tmux import refactoring. (tmux.ts file deletion deferred — Sprint
    179 follow-up; spawn_backend type narrowing partial.)

  178-006 CI flake fix — DONE
    tests/cli/archive-debt.test.ts (mock factory hygiene) +
    tests/core/orphan-cleaner-ipc.test.ts (PID liveness mock).

Auto-debt outcomes:
  178-002 (ADR-019 reconciliation) — DONE on FIX retry. ADR-019
    "Language-Agnostic Worker Verify" finally implemented (docs/adr/
    019 updated). Long-standing CRITICAL debt closed.
  178-001 (debt-175-020-fix verification gates) — NO_GO + NO_GO retry.
    Root fix is Sprint 179 W1-1 (auto-debt empty-scope inheritance);
    worker rollback (Sprint 177 Task 1) ensured src/ stayed clean
    despite repeated empty-scope dispatch.

Brain dispatch behavior observations (Sprint 179 backlog candidates):
  - FIX retry result not propagated to main task verdict in event
    stream (sequence 35-83 showed 22 minutes of stale dependency
    states); CLI aggregate summary correctly shows DONE for
    178-002/005/007 even though sprint-state.json kept original
    NO_GO. User flagged: "ana worker NO_GO ama fix DONE = doğru akış
    iletişim kanalları."
  - Brain re-evaluates DONE tasks during FIX phase (DONE → fix retry
    spawn). User confirmed this is intentional ("Brain done işleri
    tekrar dürüstçe değerlendiriyor, dokunmayalım").
  - DIRECTIVES Dependencies field plan-slot vs disk-task-ID mapping
    drift when auto-debt prepends (178-005 misrouted to auto-debt
    178-002 instead of plan-slot Doc updates).

Verdict: 9 DONE + 0 debt + 2 NO_GO → CLI shows full GO assessment
(verdict matrix per DIRECTIVES would normally call this NO_GO due
to "Task 5 fails outright" but the FIX retry closed 178-007). Sprint
verdict acceptance per Brain auto-evaluation.

Crisis Stabilization §4 closed. Sprint 179 (sub-project #2 original
12 task + 1 Bug A fix) ready to spec.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…Bug A foundation

Sprint 179 deliverables (modified files; new src/ files lost due to uncommitted state — recovery in Sprint 181):
- Bug A foundation: aggregate verdict + DEPENDENCY_RESOLVED_BY_FIX event channel + planDispatch aggregate-aware + buildDependenciesBlock dual digest (src/orchestra/result-evaluator.ts, event-stream.ts, result-collector.ts, prompt-god-template.ts)
- Self-security W4-W5 modifications: audit.ts, auth-provider.ts, session-manager.ts, types.ts, ws-gateway.ts (Sprint 179 hooks live, but interceptor src/api/terminal/{audit-integrity,command-guard,prompt-guard,outbound-limiter}.ts + src/cli/commands/audit-verify.ts deleted — Sprint 181 recovery)
- Planner state-hygiene W1-W3: sprint-planner.ts (auto-debt scope inherit + orphan cleanup), coverage hard-floor/aspirational split, doctor Memory-V2 cascade, dashboard TranslatorProp split, DEP0190 shell:true fix
- Quality assessor coverage parse (Sprint 180 W4-1)

Sprint 180 deliverables (survivors):
- Nervous Faz 1 partial: ipc-queue.ts (W2-2) + 7 nervous tests (bootstrap, action-handlers, integration-runtime, ipc-queue, directives-protection-auto-restore, faz1-smoke, accept-panic)
- Worker .result coverage zorunluluk (W4-1): worker-verify.ts vitest --coverage parse
- npm publish v1.0.0-beta.1 readiness scripts/validate-publish.mjs
- Nervous user guide kısa giriş: docs/guide/nervous-system.md
- OSS GA docs: README, installation, quickstart, getting-started

Critical incident: Sprint 179 NOT committed before Sprint 180 launch.
Worker-rollback.ts (Sprint 177 deliverable) git stash --include-untracked + git stash drop
cycle deleted Sprint 179's uncommitted NEW src/ files (5 self-security + 2 nervous core).
Modified tracked files survived. dist/ build intact (runtime functional).
Root cause memory: feedback_post_sprint_commit_mandatory (NEW, to add).

Recovery plan: Sprint 181 — TDD re-write of lost deliverables from
docs/superpowers/plans/2026-05-21-sub-project-2.md (sub-project #2 plan) +
dist/ runtime behavior reference + Sprint 180 NO_GO/GWT closure.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…start + worker-rollback fix

Sprint 180 GO_WITH_GATE_FAILURE root cause analysis:
- 4 DONE / 8 GWT / 8 NO_GO verdict
- 5 TS2307 errors (audit-integrity, command-guard, prompt-guard, outbound-limiter, audit-verify)
- 7 src/ files deleted by worker-rollback git stash --include-untracked + drop cycle
- Sprint 179 NOT committed before Sprint 180 launch → uncommitted deliverables lost

Sprint 181 scope (16 task, 6 wave):
- W0 P0: worker-rollback untracked-safe (scope-bounded stash + archive + pre-spawn guard)
- W1 (5): self-security recovery — audit-integrity, command-guard, prompt-guard, outbound-limiter, audit-verify
- W2 (2): nervous core recovery — bootstrap, action-handlers
- W3 (5): Sprint 180 NO_GO/GWT closure — state-tracker, smoke config, gate fix, npm publish, integration
- W4 (1): NERVOUS-TODO.md restore from mirror
- W5 (2): beta launch smoke v1.0.0-beta.1 + Sprint 181 retro + Sprint 182 stub

Locked decisions:
- W0 SEQUENTIAL FIRST blocker (W1+ ancak W0 DONE sonrası)
- dist/ runtime reference KORUNUR (npm run build only at W5-1)
- Recovery 3-source: dist/.js + sub-project #2 plan §Task NN + Sprint 179/180 retro learning
- memory.db schema ALTER YOK (Sprint 179'da uygulanmış intact)
- Post-sprint commit ZORUNLU (feedback_post_sprint_commit_mandatory)

Beta gate: June 1 2026, ~10 gün, Sprint 181 son beta-blocker.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…er-rollback untracked-safe fix

Manual recovery (Claude Code) after Sprint 180 worker-rollback deleted 7 uncommitted
src/ files via git stash --include-untracked + drop cycle. Source: dist/ runtime build
+ sub-project #2 plan + Sprint 179/180 retro learning.

Recovered src/ files (7):
- src/api/terminal/audit-integrity.ts (HMAC chain, I4)
- src/api/terminal/command-guard.ts (default-deny remote, I3)
- src/api/terminal/prompt-guard.ts (input pattern, I1+I2)
- src/api/terminal/outbound-limiter.ts (tenant quota, I5)
- src/cli/commands/audit-verify.ts (CLI tamper detect)
- src/nervous/bootstrap.ts (createNervousSystemIfEnabled)
- src/nervous/action-handlers.ts (4 MVP action handlers)
- src/orchestra/sprint-state-tracker.ts (Sprint 180 W1-1)

Worker-rollback Sprint 181 untracked-safe fix (src/agents/worker-rollback.ts +99 LoC):
- snapshotWorkerScope: scope-bounded git stash via pathspec (no more sweep of
  out-of-scope untracked files)
- Pre-spawn guard: collectOutOfScopeUntracked + onWarn callback
- dropWorkerSnapshot: archive folder write before drop (.deckent/worker-rollback-history/
  {sprintId}/{taskId}/stash-{iso}.patch, 7-sprint TTL)
- NOSTASH sentinel: empty-scope stash skip
- rollbackWorkerScope: per-path checkout iteration (untracked-tolerant)

Test recovery (5 + 1 new untracked-safety + 1 sprint-state-tracker):
- tests/api/terminal/audit-integrity.test.ts (5 case, I4 invariant)
- tests/security/prompt-guard.test.ts (6 case, I1+I2)
- tests/security/command-guard.test.ts (12 case, I3)
- tests/security/outbound-limiter.test.ts (4 case, I5)
- tests/orchestra/dependency-aggregate-fix-aware.test.ts (5 case, Bug A foundation)
- tests/agents/worker-rollback-untracked-safety.test.ts (5 case, Sprint 181 P0)
- tests/orchestra/sprint-state-snapshot.test.ts (3 case, Sprint 180 W1-1)

NERVOUS-TODO.md restored from /home/alperen/.claude/plans mirror (540 lines, 31KB).

DIRECTIVES.md: Sprint 181 deferred for manual recovery; new Sprint 181 (post-recovery
continuation work) will be specced after this commit lands.

tsc --noEmit: exit 0. vitest sweep: 16785/16845 pass (5 pre-existing failures in
Docker e2e + cli/run, scope-out).

Memories: feedback_post_sprint_commit_mandatory, project_worker_rollback_untracked_bug.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

`.cursor/` `.codex/` `.gemini/` adapter audit dokümanına kapanış kaydı:
- Commit Geçmişi tablosu (3a2f06e, e45ffbe, 5646520, 685e65c, 9a353a4)
- Doğrulama özeti güncellendi
- Kapanış bölümü: 7 sorundan 6'sı düzeltildi, kalan maddeler izleniyor
- Gelecek Öneriler #2'ye "Part 2" (DECKENT.md agent-rolleri redesign'ı —
  Alperen kararıyla ertelendi) notu
- README indeks açıklaması güncellendi

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…status anchor + queue preview

Closes the orientation findings (Alperen live test):
- #1 visible cursor: InputBar renders an inverse-video caret → you always SEE
  where you are. Reuses the tested editInput reducer + InputHistory.
- #4 arrow keys: ←/→/Home/End move the cursor, ↑/↓ walk history, Backspace/
  Delete/Ctrl-A/E/U edit — full line editing (was append-only).
- #2 'working vs done': a PERSISTENT status anchor above the input shows the
  phase every frame — ⠋ düşünüyor… / üretiliyor… (animated, so a mid-reply
  network pause still reads as 'working') / ✓ hazır · sıra sende when idle.
- #3 queue visibility: queued continuation prompts render as '⋯ kuyrukta N: …'
  previews while busy (was invisible).

InputBar owns input only while active (isActive=false during the confirm modal);
batched/pasted Enter is split so completed lines submit. PTY-harness verified
(scripts/ink-pty-test.mjs now supports <LEFT>/<CR>/<BS> tokens): edit + reply +
clean exit code 0. i18n: tui.generating/ready keys. Opt-in DECKENT_INK=1.
tsc+build+tests/cli green.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

… starts immediately)

Sprint start ran the FULL vitest suite TWICE before SPAWN (~400s combined),
blocking start with no user value (block_on_test_fail=false → didn't even gate):
 1. captureVitestBaseline (sprint-controller Phase 1.9, 180s) — honesty baseline.
 2. ci-guardian pre-sprint vitest (runPreSprintValidation, full suite, the
    "vitest had N failures" warning).

New `pre_sprint_tests` config flag (default FALSE) gates #1; sprint-phases now
passes `{track_test_count:false}` to runPreSprintValidation (uses its existing
configOverride param) to skip #2 while KEEPING the fast tsc gate. Both default-off
→ sprint starts immediately. Honesty verify-delta degrades gracefully without a
baseline; opt back in with `pre_sprint_tests: true`.

Zero test churn (new flag default-false + existing configOverride; no
DEFAULT_CI_GUARDIAN_CONFIG change). 215 config/ci/ci-pre-sprint tests green; tsc
clean. Surfaced live during the Sprint 255 dogfood ("sprintler çok geç başlıyor").

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

… live engine (gaps #1+#2)

Closes the top-2 dormant seams from docs/analysis/deckent-capability-maturity.md §5:

Seam #1 — reenqueueRecurring → runtime loop (AUT-5 end-to-end):
- backlog.ts: queryDue now surfaces pending recurring entries (cadence is
  gated at flip-time by reenqueueRecurring; pending recurring = due now);
  new applyRecurringReenqueue persists the done→pending flip atomically
  only when something changed.
- runtime-loop.ts: buildEngineRuntime's backlog loader applies the
  recurring re-enqueue on every tick — a recurring entry now fires at
  each cron cadence instead of dying after its first run.
- CLI/MCP user surface: `deckent autonomous backlog add --cron <expr>`
  + MCP `cron` param create recurring entries; malformed cron rejected
  at intake via core nextRun (i18n en/tr: autonomous.backlog.invalid_cron).

Seam #2 — makeWorkGeneratorSource → trigger composition (AUT-9 live):
- runtime-loop.ts: optional generateWork producer composes a
  work-generator trigger source at the LOWEST priority; candidates are
  enqueued into the backlog first (execute-dispatcher's status writeback
  requires the entry to exist) via new backlog.enqueueCandidates
  (any-status id dedupe, validation, fail-safe).
- work-generator-source.ts: makeDebtWorkGenerator — production producer
  mapping ACTIVE Memory-V2 debt records to candidates (HIGH/CRITICAL →
  risk-tagged), SQLite scans throttled to work_generator.interval_ms.
- Flag-gated config: autonomous.work_generator { enabled: false,
  interval_ms: 600000 } + validation; CLI handleStart wires the producer
  only when enabled (backward-safe, default-off).

Honest gap left explicit: todoMarkers half of WorkGeneratorInput has no
production scanner yet — debt records are the only live feed.

TDD: 19 new tests (engine-wiring 6, backlog 7, debt-generator 5 + cron
CLI 3, config 4); tsc clean; 285 targeted tests green.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…dy-read verified

- 040 [BOTH]: S279/280 behavior evolution recorded — approve-mode 10s panic-gate
  timeout (SAFETY_FLOOR exempt), edit end-to-end live (modifiedPayload), cross-
  process approval round-trip; detectors 5→12; safety-floor/presets/30-actions verbatim ✓
- 041 [BOTH]: taxonomy verified (15 agents, no test-writer, 0.40 threshold);
  distribution-goal reality: chronic refactorer relapse → ADR-072/075 mitigations
- 042 [BOTH]: task-mode now the autonomous engine's execution primitive (5 live
  consumers); THIRD MODE direction recorded — process (ADR-067 proposed; Alperen:
  'sprint' not universal + task-mode not agentic); style≠surface (ADR-081 REPL)
- 043 [BOTH]: 3 layers re-verified (boot-wired crash handlers, atomic checkpoint,
  restore); battle-tested (S267 sleep-crash 6/6, S270 WSL-VM crash)
- 044 [BOTH]: 4 call-sites re-verified (current lines); known gap — dashboard
  .dashboard snapshot stale post-finalize (UX-audit finding #2, product-sprint)
- 045 [BOTH]: dogfood FLIP false→true LIVE-PROVEN (S279/280 multi-wave);
  dependency-satisfying set = DONE ∪ MANUAL_REVIEW_REQUIRED (S280 9f966ee);
  DECKENT.md:52 stale 'deckent-dev false' claim fixed

memory.db synced, exports regenerated.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…dy-read verified

- 062 [BOTH]: sub-project #2 DELIVERED (4 guard modules: command/prompt-guard,
  outbound-limiter, audit-integrity); security invariants verbatim (AUTH_DISABLED
  deliberately ignored + timingSafeEqual); @lydell/node-pty rename; LIVE proof
  (UX-audit token auto-mint); known UI overlap bug → product-sprint
- 063 [BOTH]: consent pattern became reusable anchor (S270 F1-IMG --fix-image
  cites this ADR); 23 tests verbatim; numbering gaps explained (reserved slots)
- 064 [BOTH]: behavioral contract LIVE (dispatchTick + LEGACY_FIFO escape,
  S279/280 proof) BUT planner-bypass found — planDispatch pure-planner is
  tested-but-unwired (dispatchTick doesn't call it; :180 comment wrong) → ADR-064-W
- 065 [BOTH]: sync-to-product verified; audits-immutability is registry-exclusion
  (not literal path-guard — never register docs/audits); axis clarification vs
  MOD-SPLIT (develop/product ≠ community/enterprise); active migration mechanism
- 066 [BOTH]: resolver lives in task-types.ts; INVARIANT-DRIFT ?? 'claude' 3→9
  → ADR-066-W; S248-254 evolution recorded (host-adapters + OAuth mounts +
  ProviderCommandSpec + billing-follows-auth supersede the API-key-only auth table)

MASTER-PLAN: + ADR-064-W (planDispatch wire), ADR-066-W (claude-fallback re-audit).
memory.db synced, exports regenerated.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…rame throw (SP-1 M2p2a review)

Final-review findings #2 (collision-free synthesized ids for id-omitting OpenAI-compat backends) + #4 (Anthropic mid-stream error frame must throw, not silently complete as success — CLAUDE.md no-silent-failure).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…r persist (SP-1 M2p2b review)

Opus combined-review follow-ups: #1 cancel() breaks the per-call loop so a later auto-tier tool in the same batch cannot run; #2 a self-modifying-elevated call never persists a session/always grant (each deckent-source write re-confirms — the guard's intent is preserved). Tests: elevated-non-persist, cancel-stops-later-auto-tool, ordering-B (requestPermission parks first).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…d assertion (M3 review)

Final-review findings: #2 dist-missing must SKIP→exit 0 (header/sibling consistency, no pre-build false-red); #3 tighten confirm-card assertion to the strict 'y = allow'/'y = izin' form.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…ommand, not a JSON edit)

JSON-edit yerine tek komut: autonomous.enabled=true'yu project config'e yazar (diğer
key'leri korur, hermetik — global config okumaz), güvenlik-sözleşmesi banner'ı basar
(RBAC→policy→risk; onay-gerektiren/risk-etiketli PARK eder; yıkıcı sessiz-çalışmaz).
Default OFF korunur (kasıtlı opt-in). autonomous.disabled mesajı da `enable` komutuna
yönlendirir. i18n en/tr, TDD 2 test, tsc temiz, 30 ilgili test yeşil.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…e [--mode]`

Autonomous'a paralel: tek komutla nervous_system.enabled=true + opsiyonel authority
preset (strict|balanced|autopilot|full-auto), diğer config key'leri korur. Default OFF
korunur, authority default 'balanced' (orta/yüksek-risk→insan-onayı, 5 safety-floor hep
açık-onay). Geçersiz preset reddedilir (enable etmez). Güvenlik-sözleşmesi banner'ı +
i18n en/tr. TDD 3 test, tsc temiz, 21 ilgili test yeşil.

make-usable batch #2 (autonomous+nervous one-command enable) TAMAM.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…oval codes, bot chat grounding

From Alperen's live dogfood feedback (nervous-on sprint-289):

#1 Stale-worker false-positive noise (was: WORKER_RESPAWN spam on DONE workers)
  - sprint-state-tracker: a worker that has written its .result is FINISHED, not
    active — its .hb is never deleted so it ages out and read as 'stale'. Exclude
    such .hb from activeWorkers at the source (every consumer benefits).
  - stale-worker detector: default threshold 3min → 10min (finished workers are
    already excluded; this is only for a genuinely-hung ACTIVE worker, so don't
    flag a long tool-call early). Configurable via stale_worker.threshold_ms.

#2 Bot chat quality (3/100): the conversational path grounds the persistent claude
   session in NOTHING but tool directives. buildBotSystemPrompt() now injects the
   live project context (.brain/exports/summary.md, bounded) + a 'be an accurate
   deckent-expert, do not make things up' instruction. Volatile state stays
   tool-driven (deckent_status), never baked into the prompt.

#3 Short approval codes: nervous notifications showed a full UUID — typing
   'approve <uuid>' on a phone is torture. Proposer now mints a deterministic
   5-char shortCode per notification; every operator surface (Telegram message,
   deckent status/watch, CLI/REPL nervous accept) shows + resolves it. Also fixed
   a real gap: connector-notify-adapter never rendered notification actions, so
   the Telegram message now carries the copy-pasteable approve/reject command.

Follow-up to f326a68: narrow isObserverNoiseFile to ONLY self-cascade files
(ERRORS.md/nervous-*/metrics/events.jsonl). .hb + sprint-state.json are written by
the worker/controller, never self-cascade, and are useful debounced triggers — so
they're no longer filtered (restores the observer-phase-guard contract).

tsc clean; nervous+connectors 575/575 green (incl. new tests for each fix).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…ent tasks

Structural floor: an economy-tier model (haiku family — haiku, gpt-5-mini,
gpt-4.1-mini, gemini-2.0-flash) may ONLY run a document-write / audit task.
A code-development task (anything touching source, incl. in-code i18n string
work) can never auto-default to economy — the router upgrades economy→standard
(provider-appropriate) unless the user explicitly pinned the model (forceModel).

Live proof (Sprint-283): the FIX-router misrouted a tsx-i18n task to haiku +
doc-writer; CC caught it pre-spawn. This guard re-asserts the floor in both
model-finalization paths.

- New shared guard: src/core/model-tier-guard.ts (enforceModelTierGuard,
  isEconomyAllowedForKind, isCodeKindString). Pure, deterministic,
  provider-agnostic via model registry; English diagnostic reasons (mechanism
  module, no i18n binding). Self-contained scope→kind classifier (no orchestra
  import — ADR-008).
- Wire #1+#3 (routing/planner default + brain model): resolveTaskModel
  (model-selector.ts) guards the auto-selected model before return. forceModel
  paths return early (Layer 0) = explicit override honored. sprint-planner.ts:414
  / decision-engine.ts:169 inherit the guard transitively (no planner-smoke edit).
- Wire #2 (FIX/reroute): mid-sprint-adapter.applyReroute re-asserts the floor
  using task.type + task.scope, honoring task.forceModel.

TDD: 16 guard-unit tests + 5 reroute tests; updated 6 stale tests that pinned
the old (buggy) economy-on-code behavior to assert the new floor (+ added
doc-scope economy-allowed and explicit-override-honored cases).

Verify: tsc --noEmit clean; tests/core + tests/orchestra green (655 files,
11242 tests, 0 fail).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

… MCP-table drift/dup

#1 ensureDeckentImport: was forcing @DECKENT.md auto-load back whenever the
literal `@DECKENT.md` was absent — which reverts a deliberate on-demand
"see DECKENT.md" reference (e.g. the CLAUDE.md context-trim) on the next
`deckent sync`. Now reference-aware: ANY mention of DECKENT.md satisfies the
requirement; only prepend the @-import when there is NO reference at all.
Backward-compatible (existing @DECKENT.md files unchanged). +1 test.

#2 DECKENT.md: the inline "MCP Tool Reference" table had drifted to 33 rows vs
the canonical 35 (missing deckent_usage + deckent_process) despite the file's
own note that the list is auto-generated. Replaced the hand-table with a pointer
to docs/reference/mcp-tools.md (single source). De-duped the agent/skill name
enumeration (already in the Built-in Agents/Skills sections) to a pointer. -38/+21.

53/53 affected sync/utils tests green; no test asserts DECKENT.md table content;
tsc --noEmit clean.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…t (B-SENTINEL-CLOBBER)

Lifecycle-robustness #2 (sprint-323). writeHonestSentinelResult wrote a
worker-crashed-no-result NO_GO sentinel whenever a .result was absent — but
absence is NOT a crash when the sprint was already finalized and the result moved
to .brain/archive. sprint-323: finalize --force archived 26 real results, then a
late honest-gate pass (orphan RETRO) overwrote them with brain-honest-gate NO_GO
sentinels, destroying the true verdicts (the committed code work survived only
because the sentinel touches .result, not src).

Fix: writeHonestSentinelResult is now a NO-OP when a real result already exists —
either in .tasks (worker wrote one) or archived under .brain/archive/sprint-*-tasks/
(sprint finalized). New archivedResultExists() scans the archive. The genuine
crash case (no result anywhere) still writes the sentinel (Scenario-c preserved).

Faithful: stub-eradication tests — sentinel does NOT clobber an existing .result
(stays DONE/w-real) nor write over an archived result; pre-fix both were
clobbered. tsc=0; tests/orchestra pass. Build deferred to end of P0 batch.

Part of LIFECYCLE-ROBUSTNESS-FIX-PLAN.md (P0-B of 3).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…n spec

Turns the confirmed #2 wiring gap into an implementable flag-gated plan
(DESIGN-ADR-075-AFFINITY-REORDER.md): reorder routeTaskV2 to skill-first so
selectBestAgent can build the affinity context, gated by routing.skill_agent_affinity
(default-off → byte-identical), with faithful test plan + routing-balance gate.
Bundles with iter-2 agent-cache finding. Implementation attended-defer.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…l matrix)

The token counter never worked because deckent captures the worker's final-text
stdout, not the provider's usage — which lives in each provider's NATIVE source, not
stdout. Gap spans MULTIPLE providers + classes (claude+codex buildArgs lack a
usage-emit flag; gemini already has --output-format json; API workers default 0/0) —
a Claude-only fix is wrong (Law #2).

Design grounded in proven references: tokscale (per-provider native session-store
extraction → one schema), LiteLLM (normalized usage + completion_cost across 100+
providers), Vercel AI SDK (rich LanguageModelUsage schema w/ cache+reasoning detail),
Hermes/OpenClaw (API side via OpenRouter unified gateway). Provider matrix in 4
usage-source classes (A session-store CLI · B HTTP-response API · C unified gateway ·
D SaaS web-API). Contract: each adapter's extractUsage reads its class's native
source; one rich normalized schema (add cacheWrite+reasoning); cost via pricing table;
gateway-first (OpenRouter) for the API side. Verified per-provider, full-matrix.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

… (sprint-328 dogfood)

Implements the design (docs/.../2026-06-26-provider-agnostic-usage-cost-design.md) so
EVERY provider surfaces real token usage from its NATIVE source — NOT a Claude special
case (Law #2). deckent dogfood, 6/6 DONE (5 DONE + 1 TECH_DEBT), tsc=0, suite green:

- token-usage.ts: rich normalized schema (added cacheWriteTokens + reasoningTokens,
  AI-SDK parity).
- Class-A CLI-agents: claude (subprocess.ts usageEmitArgs --output-format json +
  claude.ts extractUsage), codex (CODEX_USAGE_EMIT_ARGS --json + token_count parse,
  same shape as its /sessions/*.jsonl native store — tokscale pattern),
  gemini (verified --output-format json → usageMetadata incl. thoughts=reasoning).
- Class-B API: agentic-worker-entry accumulates response usage across the loop →
  result.tokenUsage (ollama/openai-compatible/bedrock; replaces zeroTokenUsage).
- Class-C gateway: openai-compatible parses OpenRouter normalized usage (cache+reasoning).

CC fix: codex.test.ts spawn assertion updated for the appended --json (the codex task
left it as TECH_DEBT). Full affected suite 7367 green.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…rk (16/16, 0 NO_GO)

Track A (sprint-330 fixes): error-convention (3 generic throws → DeckentError DECKENT_E072/E073),
opus outputTokens:null robust capture, KPI live-backfill (1067 measurements from 96 sprints).
Track B (Beta): F1-012 provider-registry de-hardcode (Law #2 PASS — unknown fails honestly),
F1-014r auth non-leak, codex token parity, Dockerfile.worker ship, B-HANDOFF/B-ZOMBIE hygiene,
KPI Faz-2 surfaces (deckent_kpi MCP tool, /api/kpi, kpi --trend).

Independent verification: GO_WITH_TECH_DEBT. Regen IDENTITY/README stats (new MCP tool count).
3 KPI bugs found for sprint-332: (1) `deckent kpi` no-arg returns sprintId:null (no fallback to
latest finalized sprint) — proof-of-function blocker; (2) forward-collection doesn't fire at
finalize (recordKpiMeasurements); (3) cost/token KPIs=0 (backfill has no per-task telemetry).
Backfill itself verified working (kpi --sprint <id> → 8 real KPIs).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…nnels + born-427/428 closed + docs

Second consecutive perfect sprint (15/15 DONE, 317/317 verify, tsc=0):
WLT emit+reader (flag-gated progress-stream) · telegram/terminal/nervous approval
channels · rollback wire-or-kill decided+completed · spawnlock TOCTOU closed ·
doc-pillars disk-verified · WM-7 language penalty (gated) · corpus-lint ·
'deckent do' golden-flow command · app approval wiring (sub-flag) · approval
config load · post-sprint smoke chain audited.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

… — sprints/ + directives/

Alperen'in archive-reorganizasyonu (202 sprint-klasörü → archive/sprints/, 173 DIRECTIVES →
archive/directives/, retro'lar+pid-üçlüsü silindi) koda bağlandı:
- Yazıcılar yeni-hedefe: task-arşivi + DIRECTIVES-arşivi + orphan-hb + orphan-pid/snapshot +
  cleanup log-arşivi (constants: ARCHIVE_SPRINTS_SUBDIR / ARCHIVE_DIRECTIVES_SUBDIR)
- Okuyucular yeni-öncelik + eski-düz-yerleşim FALLBACK (mevcut kullanıcı-projeleri kırılmaz —
  Yasa #2): archivedResultExists, sdk getSprintArchive, finalize, series-metrics.mjs
- restore-quirk test-yolları yeni-düzene

Verify: arşiv-aile testleri 11/11 + finalize/sdk/series 184-test yeşil; tsc + build temiz.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…ions wiring (prefix-stable cache)

Anthropic'in prefix-instability için sevk-ettiği bayrak: per-machine bölümler
(cwd, env, memory-paths, GIT STATUS) default system-prompt'tan ilk-user-mesajına
taşınır → system-prompt PREFIX byte-stabil kalır. GERÇEK kazanç (dürüst): CLI
worker'lar prompt-cache PAYLAŞMAZ (memory: worker_prompt_cache_finding) → bu
CROSS-WORKER cache-read DEĞİL; asıl fayda git-status'un her commit'te değişip
mid-sprint whatever-cache-applies'ı (single-session/subscription) invalidate
etmesini durdurması. Bayrak yalnız default-prompt'la çalışır (deckent hiç
--system-prompt geçmez) ve yalnız claude'da (v2.1.204 --help doğrulandı).

Wiring (Law #2 — 4 claude-arg builder tutarlı, drift yok):
  • provider-command-spec: ProviderCommandSpec.excludeDynamicPromptSectionsFlag
    (claude=flag, codex/gemini=null) + buildProviderCommand emit (docker + spec).
  • CLAUDE_SUBPROCESS_CONFIG.buildArgs + buildCommandString.
  • ClaudeProvider.buildCommand (subprocess + tmux branch).
  • tmux buildProviderWorkerCommand (claude branch).
Config: PromptConfig.exclude_dynamic_system_prompt_sections (default TRUE) +
DEFAULT_PROMPT_CONFIG + boolean-validation. Opts: ProviderSpawnOptions +
tmux SpawnOptions .excludeDynamicPromptSections; sprint-spawner her spawn-path'e
config-global değeri geçirir (adapterRouted/effectiveBackend/tmux, iki fonksiyon).

Provider-agnostik guard doğrulandı: codex/gemini opts-set olsa bile flag ASLA
emit edilmez (spec-flag=null). Test: +5 (spec claude ON/OFF + codex/gemini-null;
subprocess buildArgs/buildCommandString). tsc 0-hata, 366+144 test yeşil.
Canlı-binary smoke ayrı adımda (default-on doğrulaması).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

A fresh `deckent init` project (or root-only README+LICENSE repo) has ZERO
tracked directories, making the invented-dir rule unsatisfiable for every
nested path: git ls-files exits 0 with empty stdout, the gate still ran, and
every first-sprint src/ task hard-blocked without --force-scope (Law #2,
REPRODUCED red-first in tests).

Posture per Alperen (2026-07-10): Advisory-WARN. Predicate is structural, not
a threshold: trackedDirs.size === 0 — exactly the condition under which the
rule has no signal (extends the literal "0 tracked files" wording to root-only
repos with the same rationale; advisor-endorsed, flagged here).

- scope-gate.ts: greenfield classify -> new-plausible with honest reason;
  additive ScopeGatePass.greenfield/greenfieldNotice (only when >=1 write was
  unvalidated). Wrong-dir rule untouched (self-disabling via empty byBasename).
- sprint-controller.ts: surfaces the notice on BOTH channels per the
  sprint-finalizer advisory precedent — SCOPE_GATE_GREENFIELD_ADVISORY event
  (MCP/bot/dashboard see it) + one console.warn line (CLI).
- Bonus (same classify() fix): worker prompts in root-only repos no longer
  label every nested file "Unverified ... STOP+NO_GO" (mass-NO_GO noise);
  they label "New — expected to create". Twin reproduce test added.

Verify: scope-gate 13/13 + pcomp 14/14 (5 new tests, 2 written RED-first),
sprint-controller/debt-integration/scope suites 216/216, tsc EXIT 0.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…gistry (born-549 generalized)

Daemon side of the Electron desktop shell (approved blueprint, phase B1):

- src/api/serve-daemon-meta.ts: .deckent/serve-daemon.json handshake file —
  atomic temp+rename, mode 0600 re-asserted against umask (carries live API +
  terminal tokens, gitignored). Readers treat it as a HINT: adopt only after
  pid-ownership + /health identity verify (stale-after-crash is by design).
- server.ts: /health gains loopback-gated identity fields (version/pid/
  projectRoot/terminalEnabled) for the adopt-vs-spawn decision — remote
  callers keep the exact minimal body (no new fingerprinting surface).
  HttpApi.apiToken exposes the resolved bearer token (additive).
- serve.ts: writes the handshake on startup (fail-soft i18n warn), clears it
  on shutdown — which surfaced a REPRODUCED pre-existing bug:

FIX (advisor variant-B): every command-level process.on(SIGINT/SIGTERM)
listener is dead code — entry.ts's bootstrap onSignal wins registration order
and exits synchronously, so serve's graceful api.close() never ran on a real
signal. born-549's REPL teardown registry is generalized into cycle-free
src/cli/helpers/shutdown-hooks.ts; onSignal awaits it (bounded 5s, settle-all,
no-hook sync fast-path preserved); registerReplTeardown stays as alias (REPL
call sites + sigterm-teardown.test.ts untouched). serve registers
clear-then-close (sync clear first — a hung close can never block it).
Remaining 5 dead-listener commands (nervous/chat/flow/heartbeat/dashboard)
filed as born-587.

Verify: real-binary REPRODUCE proof (pre-fix SIGTERM left stale handshake ×2;
post-fix SIGTERM ✓ + SIGINT ✓ cleared + clean exit). 15 new tests; both signal
suites unmodified-green (4+13); serve.test.ts's 3 dead-behavior tests
rewritten intent-preserving + strengthened (negative listener assert,
clear-before-close order, clear-survives-rejection). tests/api 900/900,
targeted 49/49, tsc EXIT 0. tests/cli's 42 latent reds reproduce on clean
HEAD (pre-existing family #2, untouched).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…rd-isolation pattern)

- src/desktop/: package.json (private, electron 43 + electron-vite 4 +
  electron-builder + @playwright/test, all desktop-scoped — root publish tree
  untouched, npm pack --dry-run: 0 desktop entries), .npmrc restating the
  root ignore-scripts security posture (npm per-project config would silently
  bypass it; electron binary fetched explicitly via bootstrap:electron only
  when launching), tsconfig (bundler resolution, dashboard precedent),
  electron.vite.config.ts (ESM main / CJS sandboxed preload / thin pre-daemon
  renderer — real dashboard is loadURL'd from the daemon, never built here).
- src/desktop/src/shared/desktop-api.ts: typed preload⇄renderer contract SSOT
  (zod ConnectionProfile — 4-kind matrix schema-complete per Law #2, tokens
  deliberately never stored; DeckentDesktopApi UI-grade surface).
- Root integration: tsconfig exclude + copy-assets walk-skip (the exact
  dist-leak gotcha class), .gitignore (node_modules/out/release), scripts
  (tsc:desktop, build:desktop, dev:desktop, test:desktop, install:all chain),
  vitest.desktop.config.ts (Electron-free unit scope).

Verify: tsc:desktop EXIT 0 (electron 43.1.0 resolved), root lint green
(2091-file hermetic-lint 0 violations), npm pack isolation proven.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…ho skip (sprint-396) + lint crossReach sınıfı

- born-601 (sprint-396 T1+T2): 4 agent + 6 skill manifesti onarıldı — architecture/
  migration intent'e, database/monorepo/git stackDetection+intent'e, onboarding/
  provider-cli dar-sinyale; integration-engineer + terminal-ux-engineer $or-collapse
  (gerçek-kuralın skorunda: 8/6 — advisor: DÜZ-SİLME yabancı-projede bugün-ateşleyen
  kuralı kaybederdi, Yasa-#2). Worker dürüst-DEBT bıraktı çünkü goCriteria'm
  $or-tasarımıyla çelişiyordu (ölü-kelime kasıtlı korunur) → Brain-fix: lint'e
  "Cross-reach (sanctioned $or)" sınıfı eklendi (gerçek/alias-kardeşli $or-dalı
  kelimeleri borç değil); 3 orphan-girdisi emekli. SONUÇ: Known debt 11→0,
  8 alias-canlı + 3 sanctioned, 0 yeni ölü-kelime.
- born-603 (T3): fix-wave "no defect / no source change" debt'leri artık CRITICAL
  yeniden-doğmaz — konjonktif desen (fix-köken originTaskId VE tam-content
  no-defect işareti; 80-char-title değil), skip-without-permanent-resolve
  (denetlenebilir), enjekte-description artık debt-notunu taşır. Yapısal v2 =
  born-604 (ledger'lı).

Sprint-396: 3/3 (2 DONE + 1 dürüst-DEBT→Brain-kapattı), 0 NO_GO. Verify:
routing+planner bataryası 215/215 + yeni 50/50, vocab-lint EXIT 0 (borç 0),
npm run lint 0, tsc 0.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

… tam-uygulama)

Sprint-397 (15/15, 0 NO_GO): 8 test-stale ailesi + ELOOP DeckentError CODE-FIX +
katalog re-zero/$or-uyum + docs-site 2-blocker (vitepress build EXIT 0) +
README sprint-badge restore (K4) + update-readme-stats kalıcılık-fix'i +
secret/spawnsync baseline-refresh (advisor'ın merge-tuzağı reçetesiyle) +
validate-publish orphan-test GATES-API rewrite.

Brain post-fix'leri:
- T8 stats-sıfırı sprint'in KENDİ finalizer'ınca geri-basılmıştı (born-605
  canlı-kanıt #2) → sprint-sonrası donduruldu (61/61).
- agsk3/onboarding-ux: RCA-anlığı 396-öncesiydi — post-601 kontratına taşındı
  (domain-veya-intent beklentisi + builtin-parite); builtins 160/160.
- routing-live-diversity: 396-T1 data-engineer'ı evrensel-dominated bırakmıştı
  (migration@6+3 < migration-specialist@10, HER projede) + integration-engineer'a
  uygulanan cross-project $or-koruması eksikti → database|db|models $or@8 geri
  (Yasa-#2 foreign-reach, lint-orphan-gerekçeli) + migration@6 korunur;
  agent-rule-rewrite pin'i özel-case'le revize; 327/327 routing+builtins yeşil.

Yeni kayıt: born-617 CI-testglob-GAP (tests/{kpi,brain,e2e,audits,governance,
docker,config,...} CI'da hiç koşmuyor + 9-dosya latent-set — Faz-0-dışı, havuza).

Verify: tam-suite 29714-pass (29-fail'in 28'i CI-dışı aileler=617'ye, 1'i fix'lendi);
hedef-batarya 327/327 + builtins 160/160 + tsc 0 + lint 0 + vocab 0 + docs-build 0.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

… muafiyeti + backend-aware force-sweep

- P0 (ana-vaat FIX-kapısından deliniyordu): cascade-skip synthetic-result'ına
  cascadeSkipped:true — handleEvaluation artık skip'e dependencies:[]'li direkt-fix
  AÇMAZ (iş incelenmemiş MRR-temelinde yeniden koşmaz) + handleCrossDependencies
  skip'ten DONE-dep'lere xfix fan-out YAPMAZ (born-624 amplifikasyonu kesildi;
  NOT_DISPATCHED-muafiyeti emsali). Retry sonraki-sprint'in işi, upstream
  incelendikten sonra
- P1 (Yasa #2): forceKillLiveWorkers artık kill.ts killSingle kompozisyonunu
  kullanıyor (subprocess/docker-first + tmux-fallback + status/lock/prompt
  temizliği) — tmux-dışı backend'de sessiz no-op bitti; öldürülemeyenler
  COMPLETE-öncesi dürüst-uyarıyla yüzeye çıkıyor
- Testler: fix-kapısı muafiyeti (skip→fix-yok · normal-NO_GO→fix-var ·
  xfix-yok) + boolean-seam sweep + composition-pin güncel

Kanıt: scheduler-single-truth 8/8 + orchestra+cli+governance+core 21395/21426 + tsc-0.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

… 528 denied-honesty · 641-spawner + publish-pin senkronu

Sprint-409 (4/4 DONE, FIX-turu dahil, 13dk):
- born-527: Home/End gerçek-Ink eşlemesi + boş-history-push guard + keylog os.tmpdir/config (Yasa #2)
- born-528: confirm-denial toolSink dürüst-çıktı yolundan (denied-işaretli; 633-sözlüğüyle tutarlı)
- 641-spawner: spawn-time yeniden-atama yalnız assignedAgent yok/generic'se; spawn-fallback
  journal'lı (source:'spawn-fallback') — born-641 ailesi TAMAMEN kapandı

CI-kırmızı (d839682) kökü — ÜÇÜNCÜ workflow-pin kör-noktası (tests/workflows/):
- publish.test.ts eski publish-otoritesi şeklini pinliyordu → emekli-gerçeğe senkron
  (dry-run-only; gerçek publish yalnız release.yml; not-match pin'i run-adımına daraltıldı)
+ repl_surface.bg_turns dogfood-flag lokal-AÇIK (642 canlı-deneyim hazır)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…rçek-durum senkronu + Windows-EOF + kalıcı pin'ler

Sprint-410 (3/3 DONE, 5.5dk): worker'lar disk-verify ile DÜRÜST keşif yaptı —
üç işin çoğu ZATEN inmişti (494→sprint-377 c8e839a · 501-tabanı→sprint-390
03adf3e · 505→380-013); MASTER-PLAN satırları bayattı. Gerçek-delta:
- 410-002: EPIPE-zarafeti Windows EOF-eşdeğerine genişletildi (Yasa #2) + sınıf-ayrımı pin
- 410-001/003: lazy-split + dedup + tek-tanım kalıcı pin-testleri (drift-kapanı)
- MASTER-PLAN 494/501/505 ✅ gerçek-tarihçeyle işlendi

DERS (386-duplicate'in ⬜-varyantı): eski ⬜-satır sprint'e alınmadan ÖNCE disk-verify —
marathon-kuralına eklendi. + born-645 (pre-flight claude-probe yanlış-negatif flake) kayıtlı.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…u kapat (ADR-G-013)

Advisor-yakalaması (yük-taşıyan): S2 flag-ON'da child.stdout'u pipe'layıp on('data')
attach ediyor → flowing-Readable KENDİ loop-ref'ini stdout-EOF'a (≈worker-exit) kadar
tutar. child'ı unref ettim ama stdout'u DEĞİL → .result yazıp linger eden worker
coordinator-loop'unu yeniden pinler (tam da child.unref'in çözdüğü MOAT-2 senaryosu),
yalnız live_trace-ON'da. moat2-linger testi flag-OFF koşuyor → yapısal olarak yakalayamaz.

Fix: pipe'lı stdout'a unref (hbInterval.unref deseni — unref'li stream 'data' hâlâ
verir ama EXECUTE result-poll timer loop'u canlı tutar; kendi başına anchor etmez).
Flag-ON linger-assertion eklendi (stdout.unref çağrıldığını doğrular).

Ayrıca advisor #2 denetimi (subprocess .log tüketicileri stream-json'da kırılıyor mu):
TEMİZ — captureSubprocessOutput line-based/format-agnostik · readLogEvents isLogEvent-guard
(subprocess ZATEN []) · readModelUsageFromLog whole-parse-then-line-scan fail-safe ·
heartbeat/liveness mtime-check · extractUsage line-scan stream-json-safe. Whole-parse eden
kırılan tüketici YOK.

Kanıt: tsc 0 · subprocess-live-activity 4 + moat2-linger 8 yeşil (flag-ON unref +
flag-OFF byte-stable ikisi de).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…aemon transport (real RunFlow SSE into the Console)

The window is no longer handed over to the daemon's dashboard on
connect: the local shell stays mounted and consumes the daemon's
tokened HTTP API directly (approved SURF-4 decision #2). markDaemonActive
registers the origin (CSP/navigation allowlist) without loadURL;
connectWindow remains only as the legacy-dashboard escape until SURF-7.

- IPC session bridge: DaemonSession {profileId, origin, apiToken} pushed
  on `daemon.session` after connect (null on disconnect) + `session.get`
  pull for renderer reloads — LOCAL-renderer tier (token-bearing).
- CSP: connect-src allows the loopback port-wildcards in BOTH twins
  (index.html meta — the enforced one on file:// — and the
  buildLocalRendererCsp header): the connect flow already hard-rejects
  non-loopback plain-http (born-600), so loopback IS the boundary.
  Chromium rejects a bracketed-IPv6 wildcard → ::1 joins dynamically.
- Renderer shell (approved stack, exact-pinned: react 19.2.7,
  react-router 7.9.4 HashRouter, @tanstack/react-query 5.90.3,
  zustand 5.0.14, react-aria-components 1.19.0; ADR-D-005 rationale
  rows added): persistent 4-view nav (Console/Chat/Approval/History),
  typed api-client for BOTH contracts — RunFlow REST
  (list/get/preview/propose/decision/start/cancel) + per-flow SSE
  (named events, id:=durable sequence, ?token= + ?after= replay) and
  the SEPARATE poll-based ApprovalBroker (/api/approvals). 404 surfaces
  as the honest terminal.run_flow_v2 flag notice. SSE events land in
  the Query cache (SSE→cache) and render in the Console's live feed.
  i18n: 11 new desktop.shell.* keys en+tr; zero shell literals.
- PRODUCT BUG found by the smoke and fixed: `deckent serve` never
  bootstrapped providers — EVERY /api/run-flow/propose died with
  "No providers registered" (born-680's API twin). serve.ts now runs
  the same bootstrap convention as do/start (fail-soft, API-only mode
  unaffected). Also: bare-IPv6 connect URLs are now bracketed.

Tests: desktop 90 green (api-client Bearer/?token=/?after=/404-flag/
named-SSE pins, CSP builder pins, session.get joins the local tier).

REAL-BINARY PROOF (playwright _electron + REAL `deckent serve` daemon
in a tmp project, run_flow_v2 on): add profile → connect (adopt) →
session push → shell mounts → 4 hash-routes navigate → POST
/api/run-flow/propose (201) → the flow appears in the Console and its
durable events stream over SSE into the live feed:
1 PROPOSAL_SUBMITTED · 2 PREVIEW_STARTED · 3 PREVIEW_READY. Screenshot
captured — the done-criterion verbatim.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…kaydı (327 satır)

Alperen 7 kararı onayladı. Beşi uygulandı, ikisi validator tarafından bloke edildi —
ve bu doğru davranıştı (governance-by-construction çalışıyor).

UYGULANANLAR:
- SSOT-001 VERIFY -> DONE. Truth 1/1/1/1/1/-/-,
  `proof=master-archive-fresh-checkout-56d5406f0`. Arşiv SHA-256 header'daki beklenen
  d6a90fc0... ile birebir, dosya tracked, `git archive HEAD` ile temiz dizine çıkarılan
  kopya aynı hash'i verdi. Bu satır 318 aktif satırın 290'ını bekletiyordu.
- 4 yeni satır: RUNTIME-FLOOR-001 (450), ERROR-REGISTRY-001 (460), CONFIG-TRUTH-001 (470)
  P00'a; LIFECYCLE-VOCAB-001 (3305) P03'e. Ledger 323 -> 327.
- TRUTH-BASELINE-001 evidence: test-failure ratchet kuruldu, baseline 118/598 -> 113/539;
  codex-analysis'in 115/591 rakamının aynı günün daha eski snapshot'ı olduğu kayda geçti.

VALIDATOR'IN DURDURDUKLARI:
- READY (karar #2) owner receipt'i olmadan MÜMKÜN DEĞİL. İki katman: (a) SSOT-003'ün tek
  child'ı MASTER-CLI-SYMLINK-FLAKE-001 OPEN olduğu için SSOT-003 READY olamıyor;
  (b) o satırı READY yapınca ADMISSION_RECEIPT_MISSING — READY, satıra+G1'e özel
  owner-issued admission receipt'i (GR-YYYY-MM-DD-..., exact file manifest,
  owner=Alperen; decision=APPROVED) gerektiriyor. Receipt üretmedim; satır OPEN bırakıldı,
  READY-eligibility gerekçesi evidence'a yazıldı.
  BU, KENDİ TESPİTİMİN DÜZELTMESİDİR: "0 READY yalnız yazım boşluğu" demiştim; eksikti.
  Doğrusu READY hem yazım hem owner admission authority gerektirir.
- P0 yeniden sınıflandırması (karar #5): "246 -> ~90" tahminim YANLIŞTI. Codex'in 4 soruluk
  politikası doğru uygulanınca savunulabilir P1 adayı 56 değil 9 çıkıyor. Toplu indirme
  YAPILMADI. Sebep: dependents=0 olan satırların çoğu RECOVERY-BORN-* — gerçek koşularda
  gözlenmiş execution hataları, yani politikanın 1. sorusundan P0 kalıyorlar.
  Bulgu: P0 enflasyonu esas olarak yanlış etiketleme değil; gerçek çare owner-set P0
  BÜTÇESİ (eşzamanlı admission limiti) olmalı — bu bir owner kararı.

Ledger: 323/318/5 -> 327/321/6 · READY hâlâ 0 (admission receipt bekliyor) ·
P0 250, P1 57 -> 61.
Doğrulama: lint:master-plan OK (327 satır, 23 receipt, projeksiyonlar in-sync) · lint:link 0.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

PAZARTESI.md geçici köprüydü ("FAZ 3 bitince içeriği MASTER-PLAN'a taşınır ve
bu dosya silinir") — taşındı ve silindi. İş-takibi artık TEK SSOT: MASTER.

Emilen içerik (MASTER'da karşılığı olmayan her şey):
- SKILL-DURABILITY-001 (3310, P0): FORCED_SKILL_UNAVAILABLE — crosswalk'ta
  satırı olmayan tek FAZ-4a maddesi; diğer beşi mevcut satırlarla kapsanıyor
  (eşleme evidence'ta).
- PROD-SPAWNSYNC-ASYNC-001 (3315): Alperen karar #2 (2026-08-02)
  transkripsiyonu — üretim hot-path'i, TEST-SPAWN-001'den ayrı.
- BOT-LIFECYCLE-HONESTY-001 (3320): 4a-ek canlı bulguları — identity-guard
  recovery komutlarını bloklamasın + SIGTERM pid hijyeni.
- CLEAN-DASHBOARD-POLICY-001 (3325): clean↔build:dashboard policy çelişkisi;
  487-CLEAN-HOLD-EXIT komşu ama ayrı (fark evidence'ta).
- Stop-line: sprint-491 kanıt seti dondurma (FAZ-0 owner kararı).
- TRUTH-BASELINE-001: paket ölçümleri (564: P1 211/19 · P2 135 · P3 116 ·
  P4 96 · P6 kapalı) + CANONICAL kritik yol P6→P1→P3+P4→P2 (OQ-XV-06) +
  P1=FAZ-4a aynı-dilim kararı + Type Check maskesi dersi.
- DOCS-RELEASE-TRUTH-001: FAZ-5 managed-docs sahiplik kuralı.

Ayrıca TRUST-ANCHOR-001 X 0→1: PR #27'nin required-check koşusu (run
30846122775) trust-anchor kodu + 84-case suite ile 3 platformda geçti; merge
ancak bu bacaklar yeşilken mümkündü — ruleset'in kendisi kanıt zincirinde.

9 canlı dosyadaki PAZARTESI referansı MASTER satırlarına yönlendirildi
(README ×2, check-test-failures, 5 test dosyası + reference-drift ×2).
RECONCILIATION-WP0.md'deki anlatı referansları bilinçli bırakıldı — o dosya
versioned tarihsel kayıttır.

Kapılar: lint:master-plan OK (337/330/25), 7 suite 121 passed / 0 fail,
lint:link 0, tsc temiz.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>