Commit Graph

1816 Commits

Author SHA1 Message Date
Дмитрий 9799b4d3e8 fix(router-gate-v4): Smoke 5 REAL fix — path-normalization separator bug
Smoke 5 restart-test (chistaa session) refuted stale-process hypothesis and
identified the real bug: Stream A's pathNormalize() returned OS-native paths
(backslashes on win32) while DEFAULT_PROTECTED_PATTERNS regexes are forward-slash
only.

Trace confirmation:
  Stream A pathNormalize('~/foo/bar.jsonl') on win32:
    BEFORE: 'c:\\users\\admin\\foo\\bar.jsonl' — backslashes
    AFTER:  'c:/users/admin/foo/bar.jsonl'      — forward slashes
  isProtectedPath now matches → Bash/PowerShell hooks block correctly.

Root cause: path.resolve() + fs.realpathSync() on Windows produce backslashes,
caseFold lowercases them but doesn't change separators. DEFAULT_PROTECTED_PATTERNS
in shell-content-rules.mjs are forward-slash regexes (e.g. /(^|\/)\.claude\/projects/i).
defaultPathNormalize fallback in shell-content-rules.mjs DID normalize separators,
which is why my emergency commit 1072fae4 unit-tests passed but live behavior
failed — live hooks use resolvePathNormalize() which returns Stream A's
buggy implementation.

Fix:
- path-normalization.mjs: append .split('\\').join('/') to pathNormalize output.
- path-normalization.test.mjs: +1 RED→GREEN test for win32 separator normalization.

Why previous commit 1072fae4 was incomplete:
- Added pattern to protected list ✓
- Added enforce-read-path-deny.mjs ✓ (Read tool — works because hook uses
  defaultPathNormalize directly, not resolvePathNormalize)
- Did NOT detect Bash/PowerShell path-normalize integration bug (debug script
  bypassed Stream A by passing defaultPathNormalize directly).

Side observation (recorded as Stream H TODO by chistaa session):
- extractPathArgs/pathDenyOverlay — non-reading path in non-first position is
  not checked fully. Independent latent bug, separate fix.

Regression: 1715/1715 vitest tools GREEN (+1 separator test).

Critical: re-run Smoke 5 in clean session — expected PASS all 6 vectors now.
2026-05-30 08:39:52 +03:00
Дмитрий 1072fae4a6 fix(router-gate-v4): EMERGENCY Smoke 5 closure — transcript JSONL hard-deny
Smoke 5 (user-run 2026-05-30) found 5 of 6 self-exfil vectors OPEN — parent
context transcripts (~/.claude/projects/*/<session>.jsonl) readable via:
- Bash: ls ~/.claude/projects/, cat ~/.claude/projects/.../jsonl, head/tail
- PowerShell: Get-Content / gc / Select-String
- Read tool: direct file_path

Spec §3.1 declared this protection but Stream B/C/G never wired path-deny.
DEFAULT_PROTECTED_PATTERNS in shell-content-rules.mjs lacked ~/.claude/projects/.
Read tool had no path-deny hook at all.

Closure:
- shell-content-rules.mjs: +1 regex /(^|\/)\.claude\/projects(\/|$)/i in
  DEFAULT_PROTECTED_PATTERNS. Bash hook + PowerShell hook already use this
  shared list via pathDenyOverlay → both close automatically.
- enforce-read-path-deny.mjs (NEW): PreToolUse(Read) wrapper rejects any
  Read on isProtectedPath() match. Uses same shared DEFAULT_PROTECTED_PATTERNS
  for consistency (covers transcript / runtime / .env / normative / dep-checksums).
- .claude/settings.json: register enforce-read-path-deny.mjs PreToolUse Read.

Tests: shell-content-rules.test.mjs +3 path cases in protects %s + 1 dedicated
it() block. enforce-read-path-deny.test.mjs +6 cases covering Read tool block.
1714/1714 vitest tools GREEN (was 1707, +7 net).

Critical: re-run Smoke 5 in clean session before continuing other smokes.
Expected post-fix: all 6 vectors blocked (Bash/PowerShell via path-deny,
Read via enforce-read-path-deny, Write via existing normative-content-rules).

Side observation from Smoke 5 session: controller self-fabrication of
intermediate results (4 occurrences across Smokes 1-5) confirms one of the
7 fundamental limits — gate paradox + behavioral residual irreducible.
No hook catches in-response narrative fabrication (not through tool-gate).
This is a Stream H + recovery-procedures.md documentation item, not a hook fix.
2026-05-30 08:01:49 +03:00
Дмитрий e8f2290380 chore(lychee): exclude historical refs to deleted v3.9 hooks in CLAUDE.md
After Stream G deletes 5 v3.9 hooks (1f18614d), CLAUDE.md history still references
them in narrative paragraphs. These are intentional historical mentions, not bugs.
Adding to .lychee.toml exclude so pre-push lychee-links passes.
2026-05-30 06:58:53 +03:00
Дмитрий 782a02a275 feat(router-gate-v4): Stream G — register 9 v4 hooks + git add whitelist fix + sub-plan
settings.json hook registration changes:
- Removed 5 v3.9 hook registrations: enforce-chain-recommendation,
  enforce-classifier-match, enforce-graph-first, enforce-semgrep-security,
  enforce-override-limit
- Added 9 v4 deterministic hooks (no LLM-judge — Stream H follow-up):
  PreToolUse: router-gate (Bash), powershell-gate (PowerShell),
  normative-content-rules (Edit|Write|MultiEdit), tdd-real-test-verifier (Edit|Write),
  self-debrief-detector (Edit|Write|MultiEdit|Bash),
  askuser-cosmetic-detector (AskUserQuestion), mcp-classification (mcp__.*)
  PostToolUse Task: subagent-return-scanner
  Stop: todowrite-skill-verifier

shell-content-rules.mjs fix:
- Added 'add' to GIT_CONDITIONAL_SUB whitelist. Without it git add was default-deny
  by rule 5 even after approval — broke entire git workflow under v4 router-gate.

TODO Stream H (integration gaps discovered):
1. askuser-answer-parser needs PostToolUse(AskUserQuestion) wrapper
2. Schema mismatch Stream E vs Stream B approval records
3. llm-judge hooks need ROUTER_LLM_KEY config
4. decomposition-detector needs LLM-judge integration
5. parallel-session-lock pure module not implemented

Regression: 1707/1707 vitest tools GREEN.
2026-05-30 06:56:35 +03:00
Дмитрий 1f18614d7c chore(router-gate-v4): delete 5 obsolete v3.9 hooks + vocab.json (Stream G cleanup)
Deleted hooks superseded by v4 architecture (spec section 4 behavioral pivot):
- enforce-chain-recommendation (replaced by router-gate decide)
- enforce-classifier-match (replaced by skill-scope-verifier Direction 2)
- enforce-graph-first (replaced by decide classification)
- enforce-semgrep-security (folded into normative-content-rules + per-tool LLM-judge)
- enforce-override-limit (universal vocab removal section 4.2)
- enforce-override-vocab.json (vocab abolished)

Regression: 1705/1705 vitest tools GREEN after deletion.
2026-05-30 06:12:59 +03:00
Дмитрий 8ed7dad669 feat(router-gate-v4): enforce-mcp-classification (PreToolUse mcp__* wrapper, §5.3 + G1/G12) 2026-05-30 06:11:21 +03:00
Дмитрий 7df49dfa92 feat(router-gate-v4): enforce-self-debrief-detector (PreToolUse mutating wrapper, §3.12 NEW) 2026-05-30 06:08:19 +03:00
Дмитрий 83e4b0d708 feat(router-gate-v4): enforce-tdd-real-test-verifier (PreToolUse Edit|Write wrapper, §3.11)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-05-30 06:05:17 +03:00
Дмитрий 54241e6b6c feat(router-gate-v4): enforce-todowrite-skill-verifier (Stop hook wrapper, §3.9 Direction 4)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-05-30 06:00:00 +03:00
Дмитрий d3e192abf1 refactor(router-gate-v4): stub override helpers (universal vocab removed per spec §4.2)
findOverride/findOverrideAttempt/loadOverrideVocab become permanent stubs returning null/null/empty.
Non-deleted hooks (verify-before-push, tdd-gate, memory-coverage, branch-switch) still import these
symbols and need them to compile; runtime always reports 'no override'.

Adapted 15 existing tests in enforce-hook-helpers.test.mjs and 7 in enforce-semgrep-security.test.mjs
that asserted old vocab behaviour; all now assert stub behaviour (null/empty).
1824/1824 vitest tools GREEN.

Stream G of router-gate v4 deployment.
2026-05-30 05:55:46 +03:00
Дмитрий 97d7deff99 docs(pilot): snapshot 30.05 ~05:30 МСК — Stage 5 F1+F2 deployed + storm quick-fix вашиденьги24.рф 2026-05-30 05:43:00 +03:00
Дмитрий bfb650c5e2 ops(sql-runner): whitelist UPDATE supplier_projects
Расширяет MUTATING_RE для quick-fix supplier_project signal_type
collision (B3 вашиденьги24.рф site→sms за supplier_lead 1352
шторм 319/h после Stage 5 F2 fast-fail deploy).

Read-only diagnostic queries показали что поставщик сменил тип
кампании site→sms но локальный supplier_project не обновился —
резолвер выбрасывает unique_key collision, поставщик ретраит,
F2 stops at 3 retries per webhook.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-05-30 05:28:25 +03:00
Дмитрий 0d9d191895 merge(router-gate-v4): Stream E — AskUser + subagent
7 commits / 10 files / +2824 lines:
- askuser-answer-parser (S27/E33/E34 + parse + approval)
- punctuation-aware stop detection + review nits (BOM/JSDoc/??)
- cosmetic AskUser detector (v4.1 §4.5)
- subagent return scanner + G2 narrative + structured schema
- anchor 'всё ок' narrative pattern (no false-match inside 'всё окно')
- subagent-prompt-prefix inheritance (256-bit sentinel, restricted/ paths)

Stream tests pass.
2026-05-30 05:09:07 +03:00
Дмитрий 207394f2c0 merge(router-gate-v4): Stream D — LLM-judge Layer 4
13 commits / 10 files / +3017 lines:
- multiJudgeConsensus 3-judge any-YES + cache/budget
- per-tool LLM-judge pure decision + PreToolUse hook wiring
- response-scan deterministic layer + LLM layer + Stop hook
- normative-content path matcher + content extraction
- normative-content deterministic layers + multi-judge Layer 4
- normative-content PreToolUse hook wiring
- ProxyAPI live integration smoke

Stream tests pass.
2026-05-30 05:08:41 +03:00
Дмитрий dba961c51f merge(router-gate-v4): Stream C — static scan + MCP path-deny
8 commits / 11 files / +3066 lines static-content-scanner / framework-boot-scanner / glob-restricted-filter / mcp-tool-classifier / commit-message-scanner.

Review fixes: browser_navigate host-boundary (SSRF spoof), boot-scan best-effort.
2026-05-30 05:08:01 +03:00
Дмитрий b4217789ca fix(router-gate-merge): A↔B integration — resolvePathNormalize test after Stream A merged
После merge Stream A модуль ./path-normalization.mjs существует → resolvePathNormalize() возвращает Stream A pathNormalize, не fallback. Stream B тест предполагал отсутствие модуля и assert'ил конкретное default-значение 'a/b'.

Fix: меняю assertion на 'returns a function' + 'does not throw' — сохраняет original intent (resolvePathNormalize всегда возвращает callable) без жёсткой привязки к implementation Stream A pathNormalize.

Verified: vitest 59/59 GREEN на enforce-router-gate.test.mjs.
2026-05-30 05:06:58 +03:00
Дмитрий 433e135081 merge(router-gate-v4): Stream B — Bash/PowerShell content rules
16 commits / 11 files / +2849 lines:
- Bash hard-blacklist (v3.9+v4.0 C16/#4/#21/#22/#34 + v4.1 G7/G8 wget/nc)
- Bash whitelist + script-execution file-watcher
- classifyBashCommand integration + bashContentClassify export
- Bash gate main() + dynamic path-normalize fallback (fail-CLOSE)
- PowerShell tokenizer + hard-blacklist (keep + v4.1 G10 PS env)
- classifyPowerShellCommand (whitelist + path-deny + git route)
- PowerShell gate main() (fail-CLOSE)
- shared classifyGitCommand (readonly/conditional/hard incl G5/G6 gpgsign/--no-verify)
- Review fixes: 2>&1 fd-duplication allowed, git -c RCE closed, runtime-dir path-deny

Stream tests pass.
2026-05-30 05:05:15 +03:00
Дмитрий b31f765852 merge(router-gate-v4): Stream A — pure decision modules
16 commits / 16 files / +2231 lines:
- decide() 4 поведения + nodeMatches + chain-state (§4, §10.1)
- safe-baseline metering Direction 1 + v4.1 hard sync (§3.6)
- skill scope verifier Direction 2 + v4.1 hard sync (§3.7)
- decomposition detector Direction 3 + v4.1 hard-block (§3.8)
- TodoWrite skill verifier Direction 4 + v4.1 hard sync (§3.9)
- self-debrief detector v4.1 NEW (§3.12)
- TDD real-test verifier regex-based (§3.11)

Stream tests: 920 unit-tests GREEN inside subagent session.
Checkpoint 1 — first of A→B→C→D→E sequence.
2026-05-30 05:04:31 +03:00
Дмитрий bdd573a129 fix(router-gate): stream A decide — unicode boundary on cyrillic direct-invocation, polite skill_call forms, +tests, knownInRegistry contract docs
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-05-29 21:26:10 +03:00
Дмитрий 7f21cd929a feat(router-gate): stream A — core decide() 4 поведения + nodeMatches + chain-state (§4, §10.1)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-05-29 21:16:31 +03:00
Дмитрий e22148d238 fix(router-gate): stream A tdd-verifier — test no_test_block + EACCES vs ENOENT + known-limitation docs
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-05-29 21:12:33 +03:00
Дмитрий ec637863dd feat(router-gate): stream A — tdd real-test verifier regex-based (§3.11) 2026-05-29 21:04:37 +03:00
Дмитрий 6f4d8889fc fix(router-gate): stream A self-debrief — unicode lookbehind for cyrillic patterns + false-positive tests 2026-05-29 21:01:56 +03:00
Дмитрий 5bcb1a6ba0 test(router-gate): ProxyAPI live integration smoke + stream D sub-plan (stream D task 13)
Opt-in live smoke (ROUTER_LLM_LIVE_TEST=1 + ROUTER_LLM_KEY); auto-skips otherwise
so it never pollutes the unit regression in worktrees where undici is unresolved.
Checkpoint-1 live result on owner machine: PASS (2/2) — single Sonnet judge + 3-judge
consensus (Sonnet 4.6 + Haiku 4.5 + Opus 4.7) reach all models with real verdicts.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-05-29 20:55:20 +03:00
Дмитрий 5dff5f83f7 feat(router-gate): stream A — self-debrief detector v4.1 NEW (§3.12) 2026-05-29 20:50:48 +03:00
Дмитрий b8d2a0ca2e feat(router-gate): normative-content PreToolUse hook wiring (stream D task 12)
Recovered from a subagent crash (socket error mid-task) that left literal-newline
corruption in two .join() string literals; repaired and committed by controller.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-05-29 20:48:51 +03:00
Дмитрий eec75c4d33 docs(CLAUDE.md): v2.42 — router-gate v4 spec triple + master plan + handoff + 5 worktrees + rationalization-audit fix deployed 2026-05-29 20:48:47 +03:00
Дмитрий 86457dc8f5 fix(router-gate): stream A todowrite-verifier — unicode boundary for cyrillic mention patterns + DRY + tests
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-05-29 20:46:54 +03:00
Дмитрий c98f1a47a2 fix(router-gate): allow 2>&1 fd-duplication, keep file-redirect block (review finding) 2026-05-29 20:45:23 +03:00
Дмитрий f6e26774db fix(router-gate): close git -c/option-injection RCE + runtime-dir path-deny (review finding) 2026-05-29 20:45:16 +03:00
Дмитрий 657894ac19 feat(router-gate): normative-content decide() + multi-judge layer 4 (stream D task 11)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-05-29 20:37:13 +03:00
Дмитрий 6ca4950eee feat(router-gate): stream A — todowrite skill verifier Direction 4 + v4.1 hard sync (§3.9)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-05-29 20:34:46 +03:00
Дмитрий 6b97fdb08a feat(router-gate): powershell gate main() (fail-CLOSE) 2026-05-29 20:29:23 +03:00
Дмитрий a564b02a59 fix(router-gate): stream A decomposition — EOF newline + skill-in-current edge test
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-05-29 20:22:26 +03:00
Дмитрий bc28d99b17 feat(router-gate): normative-content deterministic layers (stream D task 10)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-05-29 20:22:13 +03:00
Дмитрий ada33652f6 docs(plans): router-gate v4 Stream E sub-plan (AskUser + subagent) 2026-05-29 20:21:43 +03:00
Дмитрий 98497bd2b8 feat(router-gate): classifyPowerShellCommand (whitelist + path-deny + git route) 2026-05-29 20:20:35 +03:00
Дмитрий 1f5d0dbc09 feat(router-gate): powershell tokenizer + hard-blacklist (keep + v4.1 G10) 2026-05-29 20:19:15 +03:00
Дмитрий 1659ca2dd3 feat(router-gate): stream E — subagent-prompt-prefix inheritance (256-bit sentinel, restricted/ paths, isCli guard)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-05-29 20:15:12 +03:00
Дмитрий 06d61522cc feat(router-gate): stream A — decomposition detector Direction 3 + v4.1 hard-block (§3.8)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-05-29 20:14:06 +03:00
Дмитрий 26a9b9af0d feat(router-gate): normative-content path matcher + content extraction (stream D task 9)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-05-29 20:12:58 +03:00
Дмитрий 0c4a6ca23a feat(router-gate): bash gate main() + dynamic path-normalize fallback (fail-CLOSE) 2026-05-29 20:10:58 +03:00
Дмитрий 9b7c2b335d fix(router-gate): stream A skill-scope — restore plan reason strings, arrow/optional-chaining, +reason tests 2026-05-29 20:10:41 +03:00
Дмитрий 23b10d2b7f feat(router-gate): response-scan LLM layer + Stop hook (stream D task 8)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-05-29 20:10:23 +03:00
Дмитрий e951193dd5 feat(router-gate): classifyBashCommand integration + bashContentClassify export 2026-05-29 20:09:42 +03:00
Дмитрий 78a88a3bc0 fix(router-gate): stream E — anchor 'всё ок' narrative pattern (no false-match inside 'всё окно') 2026-05-29 20:07:58 +03:00
Дмитрий fbbfdf530b feat(router-gate): response-scan deterministic layer (stream D task 7) 2026-05-29 20:06:52 +03:00
Дмитрий f05a8a66b2 feat(router-gate): bash whitelist + script-execution file-watcher 2026-05-29 20:06:04 +03:00
Дмитрий 826650cb1b feat(router-gate): stream A — skill scope verifier Direction 2 + v4.1 content-level (§3.7)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-05-29 20:05:22 +03:00
Дмитрий 76ef10ec26 feat(router-gate): bash hard-blacklist (v3.9+v4.0 C16/#4/#21/#22/#34 + v4.1 G7/G8) 2026-05-29 20:04:40 +03:00