@ngockhoale/ukit 3.4.81 → 3.4.83
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +30 -400
- package/package.json +1 -1
- package/src/diagnostics/failurePatterns.js +7 -4
- package/src/index/taskRouting.js +19 -1
- package/src/learning/codeProposals.js +36 -15
- package/template_project/.claude/hooks/auto-allow-bash.sh +155 -10
- package/template_project/.claude/hooks/completion-gate.sh +68 -15
- package/template_project/.claude/hooks/skill-router.sh +115 -14
- package/template_project/.claude/hooks/task-watchdog.sh +19 -3
- package/template_project/.claude/settings.json +2 -2
- package/template_project/.claude/ukit/index/reset-auto-permissions.mjs +3 -1
- package/template_project/.claude/ukit/index/route-task.mjs +27 -2
- package/template_project/.claude/ukit/runtime/execution-ledger.mjs +99 -19
- package/template_project/.claude/ukit/runtime/handoff-provenance.mjs +6 -0
- package/template_project/.claude/ukit/runtime/hook-telemetry.mjs +14 -0
- package/template_project/.claude/ukit/runtime/stop-coordinator.mjs +38 -8
package/CHANGELOG.md
CHANGED
|
@@ -2,6 +2,35 @@
|
|
|
2
2
|
|
|
3
3
|
All notable changes to UKit are documented here.
|
|
4
4
|
|
|
5
|
+
## 3.4.83 - 2026-11-13
|
|
6
|
+
|
|
7
|
+
Hook fixes from a one-week audit of this repository's own Claude Code transcripts, hook-latency rows, decision log and permission audit log (6,685 hook runs, no hook timeouts; Stop p95 about 1.6s).
|
|
8
|
+
|
|
9
|
+
- Fix: **`auto-allow-bash` no longer mints junk permission rules.** The binary was picked by whitespace splitting, so a leading assignment with a quoted or substituted value was cut in half and the next fragment became the "binary": `UKIT_EVIDENCE='… HEAD …' yarn test` minted `Bash(HEAD:*)`, `d=$(ls -td …)` minted `Bash(-td:*)`. The hook now scans real shell words (single/double quotes, `$(…)` with nesting, backticks, backslash escapes), skips whole `NAME=value` / `NAME+=value` words, and mints nothing when the first command word cannot be determined (unbalanced quote or substitution, leading operator, redirection or subshell) or starts with `-`. Behaviour change: `command -v node` and `env -i FOO=1 ls` now mint nothing (they minted `-v` / `-i`). The denylist block, charset check, interpreter list and locked atomic write are unchanged. omp runs this same script through its bridge; Codex has no auto-allow lane.
|
|
10
|
+
- Fix: `reset-auto-permissions.mjs --sanitize` also removes rules whose token starts with `-` (option fragments left by the old tokenizer), behind the same proven-provenance gate. Other junk words (`HEAD`, `status`, `view` …) cannot be told apart from real binaries and are left in place; the full reset removes every UKit-managed rule.
|
|
11
|
+
- Fix: **`task-watchdog` no longer announces "payload was truncated or stalled" on large Edit/Write calls.** Its PostToolUse path reads only the event name from the payload (tasks, state and config come from disk), yet any payload over the 64 KiB staging cap was reported as incomplete input. A payload cut only by the size cap, whose event name is provably complete and exactly `PostToolUse`, now runs normally. Stalled staging, an unrecoverable event name, Stop (which forwards the raw payload to the coordinator) and the pre-install fallback branch keep the announced degrade.
|
|
12
|
+
- Fix: **self-improve no longer mints handoff tasks for failures that never happened.** The `timeout-hang` class matched `timeout|hang|hung` anywhere in a command's text, so "c*hang*elog", "unc*hang*ed", a test file named `verifyReleaseTimeout.test.js` or an env var `UKIT_VERIFY_STEP_TIMEOUT_MS` each produced an `AUTO-<n>.md` task telling the executor to raise a timeout. Mined patterns now carry the failed receipts' exit codes (`exitCodes`); `timeout-hang` needs exit 124 or a whole-word marker ("timed out", "timeout exceeded/expired", "hangs", "hanging", "hung") on the command's invocation lines — heredoc bodies and comment lines are not read. The other three classes use whole-word matches on the same stripped text. Known limit: a timeout the host kills with exit 1 or 137 is not classified (a missed task, not a wrong one).
|
|
13
|
+
- Fix (tests only): hook latency telemetry from test runs no longer lands in the repository's own `.ukit/storage/cache/hook-latency/` (seven test files spawned hooks against the real root; about 4,200 rows over five days). `appendTelemetryRow` drops rows for the root named by `UKIT_TELEMETRY_DENY_ROOT`, which only this repository's `vitest.config.js` sets; unset, it is inert.
|
|
14
|
+
- Fix (pre-release review): the new scanner gives up — mints nothing — on constructs whose end it cannot prove. Two review rounds each found a command resolving to the wrong binary (`X=$'\' curl ' ls` and `A=${x:- curl } ls` both minted `Bash(curl:*)` while bash runs `ls`). Every `$` now goes through one helper that accepts only `$(…)`, `$NAME` and a simple `${NAME}`; ANSI-C/locale quoting (`$'…'`, `$"…"`), any other `${…}`, `$[…]`, and a `#`, heredoc or `case` inside `$(…)` mint nothing. The scan is bounded to the first 4,096 characters of the command (a 100 KB assignment value stalled the hook for more than 45 seconds; a first command word not resolved inside the window mints nothing), and variable names are matched against explicit ASCII classes (`É=a ls` no longer counts as an assignment).
|
|
15
|
+
- Tests: `tests/hooks/autoAllowBinaryResolution.test.js` (41), four size-truncation cases in `tests/hooks/taskWatchdog.test.js`, classifier and `exitCodes` cases in `tests/core/codeProposals.test.js` / `tests/core/failurePatterns.test.js`.
|
|
16
|
+
- Packaging: the 2.5.1–2.6.11 changelog entries moved verbatim to `docs/archive/CHANGELOG-2.5.1-2.6.11.md` (repository only); size limits unchanged.
|
|
17
|
+
- Not changed: decision-plane shadow calls to the local `unic-decision` endpoint timed out 184 times in the audited week and five decision keys never returned a valid answer — shadow-only, not root-caused in this release.
|
|
18
|
+
|
|
19
|
+
## 3.4.82 - 2026-11-10
|
|
20
|
+
|
|
21
|
+
Completion-gate fixes from a field-telemetry audit of the projects running UKit (Stop/gate logs, hook latency, route state, session transcripts).
|
|
22
|
+
|
|
23
|
+
- Fix: **a stale legacy handoff cursor no longer marks every stop BLOCKED.** A `docs/AI_HANDOFF/RUN.md` written before `RunId` existed has no authority record, so nobody can own it until `/ukit:handoff-fullstack --adopt-legacy-run=<RunId>`. A session whose published route holds no handoff binding is now `NOT_OWNED` (an ordinary session beside an old cursor) instead of `UNKNOWN`; telemetry showed ~2,500 "handoff ownership UNKNOWN … report BLOCKED" advisories, 2,000 of them in one project in one month. A route that carries a `handoffRunId`, a duplicated `RunId`, or a missing/foreign/incomplete route still fails closed as `UNKNOWN`; the cursor is never touched and adoption is unchanged.
|
|
24
|
+
- Fix: **questions and acknowledgements are no longer routed as change orders.** Tone-folding turned "cái" into the verb *cài* (install) — likewise chính/chỉnh, Việt/viết, đợi·đối·đội/đổi, cháy·chảy/chạy, sữa/sửa, thèm/thêm — so "Cái này là gì, tại sao có chế độ này" owed write + verification evidence. Also released: "chưa được cấp quyền cập nhật …" / "no permission to update" (a read-only banner, not an order), "còn cần làm gì thêm nữa không" / "có thêm test nữa không" ("anything else", not "add") — only in the closed "còn/có … gì thêm [nữa] [không]" / clause-final "thêm nữa không" shape, so "thêm gì vào README.md" stays an order, a what-is question followed by punctuation, a leading "giải thích …" / "explain …", and bare acknowledgements after a task ("ok", "có", "được", "ừ", "được rồi") which now park the previous contract instead of re-arming it. Every release keeps its order control: "Cái này là gì, sửa giúp tôi", "thêm test nữa đi", "ok sửa đi", "giải thích rồi sửa luôn", "cấp quyền cập nhật cho user admin trong file roles.js" still block an empty stop. A bare "yes / ok / có / được / ừ / đồng ý" that answers a question the agent itself just asked (the last assistant text in the host transcript ends with a question, with no other owner turn in between) is the go-ahead for that same task and keeps its contract armed; "đóng nhé" (close it) is no longer read as "đồng ý". The homograph / banner / question rules are applied to `src/index/taskRouting.js`, `route-task.mjs` and the `skill-router.sh` hook; the acknowledgement rules live in `route-task.mjs` and the hook (the `src/index` copy has no greeting lane).
|
|
25
|
+
- Improved: **an unproven verification now names the command that proves it.** Most verification-evidence blocks were a runner piped into `tail`/`grep`/`tee` or followed by `; echo exit=$?`, and sessions repeated the same masked shape. The PostToolUse notice and the gate instruction now add "Run exactly this once: …" (`set -o pipefail; <same command>` for a pipe, the command up to and including the runner for a status echo, keeping a leading `cd`/env prefix) — only when the same classifier accepts that command and no second runner would be dropped. Guidance only: what counts as proof is unchanged, and loop/background shapes keep the generic wording.
|
|
26
|
+
- Fix: **"semantic contract invariant violation … contradictory contracts" no longer blocks a healthy session.** Telemetry: 172 such Stop blocks. The router re-scores the execution mode on every tool event while the contract stays the task's own, and both the router and the gate re-derived the contract from the *current* mode — so a contract built under `find-cause`/`map-impact` compared unequal to itself once the mode moved to `local-fix`. The comparison now accepts a stored contract that any execution mode reproduces for the same prompt and engine; a contract no mode reproduces (foreign evidence tag, different task kind) still blocks. Two feeders are closed as well: a new task no longer inherits the previous task's evidence tags, operation goals or state targets (a read-only review's `impact-evidence` leaking into the next change order), and a system-injected turn (compact summary, `/loop` wakeup) after an answer-only turn no longer mints a contract bound to the owner's prompt.
|
|
27
|
+
- Fix: **a slow Stop evaluation is no longer reported as a broken install.** One evaluation repeated the same git probes (16 spawns → 10; about 950ms → 550ms measured on this repo); they are now memoized for that single evaluation only. The evaluator's time slice scales with the hook deadline (1500ms → up to 3000ms; `completion-gate.sh` default deadline 3000 → 5000ms, inside the hook's slot, see below), and an evaluator that still runs out of time blocks fail-closed with its own fixed text ("ran out of time … Stop again to re-evaluate") instead of "Run: ukit install".
|
|
28
|
+
- Fix: the Stop message for an unproven verification cut the command at 80 characters — on a real `cd <root> && PYTHONPATH=. venv/bin/python3.12 -m pytest … | grep … | tail -40` that ended exactly at `python3.12 -`, naming a command that never ran and hiding the pipe that made it unproven. The label now keeps the whole command up to 240 characters (both ends beyond that). The hook-deadline block now says to stop again first and reach for `ukit install` only if it repeats; it stays fail-closed.
|
|
29
|
+
- Fix: a Stop on which the completion evaluator could not finish (killed at its time slice, hook deadline, evaluator crash) blocked every time with no cap — one field session took 170+ such blocks in a row on a loaded machine, each costing a turn the agent could do nothing with. These blocks now have their own bounded breaker in `completion-gate.sh`: counted per session (one session never releases another), released only after more than `stopGateMaxCrashStreaks` (default 3) consecutive failures (no more than 10 minutes apart) that also span at least 120 seconds — stopping again a few times in a row never releases — and the release is announced as a blocked, non-complete outcome ("completion was NOT verified … not a pass"). Any real verdict resets the count; evidence blocks, and any decision carrying a handoff-cursor or task-watchdog block, are never counted. The crash breaker for a coordinator that exits non-zero is unchanged and separate. A block verdict the evaluator had already written when its time-slice kill landed is now honoured as that verdict instead of being reported as an evaluator failure (a cut-short release is still never trusted).
|
|
30
|
+
- Fix: the Stop hook's chain slot is 12s (was 8s) and its host timeout 22s (was 18s), keeping the chain-budget rule host ≥ slot + 10s. With the 5000ms coordinator deadline plus the 1500ms killer the wrapper could need 6.5s before bash start-up and staging, and a chain-runner timeout of this wrapper would have been a silent release.
|
|
31
|
+
- Tests: `tests/hooks/semanticContractModeChurn.test.js`, `tests/hooks/stopEvaluatorBudget.test.js`, `tests/hooks/telemetryAuditRouting.test.js` (real hook + gate, fresh and after-task, with order controls), a legacy-cursor bystander case with fail-closed controls in `handoffProvenance.test.js`, and rerun-guidance cases in `stopGateReceiptCapture.test.js`.
|
|
32
|
+
- Not changed (by design, per the audit): pipe/`; echo $?` shapes stay unproven; a carried compact-summary contract still owes its evidence; the teammate-message lane has no fix in this release.
|
|
33
|
+
|
|
5
34
|
## 3.4.81 - 2026-11-08
|
|
6
35
|
|
|
7
36
|
- New: **automatic Claude Code session discovery for UKit Advisor.** `ukit:advisor` / `$ukit-advisor` with nothing after the trigger (or only a short remark such as "lỗi rồi") reads the latest main-agent Claude Code session of the current project — no paste needed; `session <id>` and `project <path>` name one explicitly; anything else is still pasted output, as in 3.4.80. A bundled read-only Node reader (`scripts/claude-session-reader.mjs` inside the skill, also `ukit advisor context [--last|--session <id>|--project <path>] [--json]`) matches the project by the session's recorded cwd (realpath, git root, linked worktrees; nested repos and sibling folders excluded), honours `CLAUDE_CONFIG_DIR`, skips subagent/sidechain transcripts and symlinks, refuses to guess between concurrent sessions, streams huge/partial/concurrently-written JSONL with bounded secret redaction, and renders a bounded, redacted, untrusted digest that keeps owner messages, tool results, assistant claims and Stop-hook/gate refusals apart (owner STOP ≠ hook refusal). It never writes, spawns, resumes or messages Claude Code; no session, no permission or a Codex model that cannot run shell commands yields one safe status-request prompt.
|
|
@@ -1166,405 +1195,6 @@ OMP fully autonomous / unattended mode — cycle C35 (TASK-008..014).
|
|
|
1166
1195
|
note); opencode `bash."*": "ask"` flip deferred pending deny-set
|
|
1167
1196
|
expansion.
|
|
1168
1197
|
|
|
1169
|
-
## 2.6.11 - 2026-09-20
|
|
1170
|
-
|
|
1171
|
-
Self-improvement roadmap phases 2–4 — cycle C32 (TASK-227..232).
|
|
1172
|
-
|
|
1173
|
-
- Feedback capture: `src/diagnostics/feedbackEvents.js` + shared
|
|
1174
|
-
`ledgerFiles.js` ledger helper record labeled wrong-route /
|
|
1175
|
-
manual-correction events; new `ukit feedback` CLI.
|
|
1176
|
-
- Telemetry emitters: `skillIds` in route-audit entries
|
|
1177
|
-
(`route-task.mjs`); retriever lane stats appended to
|
|
1178
|
-
`.ukit/storage/cache/retriever-lanes.jsonl` (`retriever.js`).
|
|
1179
|
-
- Metrics surfaces: `skillAccuracy.js` + `laneStats.js` collectors;
|
|
1180
|
-
`ukit metrics` gains `feedback`, `skills`, `retriever-lanes`, and
|
|
1181
|
-
`learning` sections (lazy-import, degrade to `n/a`).
|
|
1182
|
-
- Memory promotion lanes: `ukit memory learn` mines failure-patterns into
|
|
1183
|
-
pending `ukit memory propose` candidates (`src/learning/patternProposals.js`);
|
|
1184
|
-
`ukit memory promote` writes approved project_rule/procedure records to
|
|
1185
|
-
MEMORY.md `## ukit-learned`; `ukit memory episode` writes a session-episode
|
|
1186
|
-
record (`--session`/`UKIT_SESSION_ID`/latest-ledger resolution) plus
|
|
1187
|
-
SessionEnd hook `session-episode.sh`.
|
|
1188
|
-
- Tuning loop: `learning.*` config namespace (validated in runtimeConfig,
|
|
1189
|
-
template parity) + `src/learning/tuning.js` advisory lane-weight/threshold
|
|
1190
|
-
suggestions.
|
|
1191
|
-
- Verification: per-task suites green on rerun; `.claude`↔`templates/.claude`
|
|
1192
|
-
parity diff-clean. 6/6 tasks approved, 0 fix rounds.
|
|
1193
|
-
|
|
1194
|
-
## 2.6.10 - 2026-09-20
|
|
1195
|
-
|
|
1196
|
-
Self-improvement roadmap phase 1 — cycle C31 (HK-401, SI-101..103).
|
|
1197
|
-
|
|
1198
|
-
- Hook contract fix: centralized `ukit_emit_permission_decision` in
|
|
1199
|
-
hook-input.sh (keyed on `UKIT_HOOK_CHAIN_RUNNER`); bypassPermissions probe
|
|
1200
|
-
replaced with locked fallback deny + exit 0 — dead `ask` path closed.
|
|
1201
|
-
- Router v2-memory lane: `listMemoryItems` reads memory records.json;
|
|
1202
|
-
`record` branch added to buildMemorySegments.
|
|
1203
|
-
- Outcome join: `routeOutcomes.js` joins route-audit × exec-ledger on
|
|
1204
|
-
requestKey; new `ukit metrics` command surfaces route accuracy.
|
|
1205
|
-
- Failure mining: `failurePatterns.js` mines exec-ledger into
|
|
1206
|
-
failure-patterns.json (advisory, never-throws).
|
|
1207
|
-
- Self-improvement surface audit + roadmap spec:
|
|
1208
|
-
`docs/research/2026-09-20_self-improvement-surface-audit.md`.
|
|
1209
|
-
|
|
1210
|
-
## 2.6.9 - 2026-09-19
|
|
1211
|
-
|
|
1212
|
-
Code Intelligence scale release — cycle C30 (CI-301..305, TASK-218..222).
|
|
1213
|
-
|
|
1214
|
-
- Vector lane: dep-free `vectorProvider.js` (FNV-1a feature-hashed embeddings,
|
|
1215
|
-
L2-normalized, cosine-ranked; injectable loader seam for real embeddings,
|
|
1216
|
-
degrade to hashed/null); wired as RRF lane w=0.6 between bm25 and semantic.
|
|
1217
|
-
- Analogy lane: `analogy.js` fingerprints memory `procedure` records + index
|
|
1218
|
-
symbols/files; `analogy` compile mode ranks by Jaccard, hits land in
|
|
1219
|
-
packet.memory + L2 evidence.
|
|
1220
|
-
- Co-change mining: `cochange.js` mines `git log --name-only` (spawnSync,
|
|
1221
|
-
bounded, core.quotePath=false) into `cochange.json`; impact sets gain a
|
|
1222
|
-
provenance-preserving co-change lane (kind:'cochange', confidence=count/10).
|
|
1223
|
-
- L3 summaries: `summaries.js` extractive docblock/symbol summaries for
|
|
1224
|
-
deep_flow+analogy packets; `packetToText` renders `summary:` lines.
|
|
1225
|
-
- Graph store: `graph.js` `codegraph.json` JsonGraphStore (imports+calls+
|
|
1226
|
-
cochange → nodes/edges, content-gated atomic write), contract-only
|
|
1227
|
-
scip/pdg/lsp stubs via `createGraphProvider`.
|
|
1228
|
-
- Derived artifacts wired into `ukit index build|refresh`, stale-refresh,
|
|
1229
|
-
install, and index scripts via `buildDerivedIndexArtifacts` (config-gated,
|
|
1230
|
-
never fatal, outside the core build's atomic-rename contract).
|
|
1231
|
-
- Config: `codeIntel.{embedding,analogy,cochange,summaries,graph}` +
|
|
1232
|
-
`retriever.weights.vector` validated in runtimeConfig with template parity;
|
|
1233
|
-
`enabled:false` honored for analogy/graph lanes.
|
|
1234
|
-
- Verification: full suite 2720/2720 green; unpacked ceiling 6.9MB.
|
|
1235
|
-
|
|
1236
|
-
## 2.6.8 - 2026-09-19
|
|
1237
|
-
|
|
1238
|
-
Code Intelligence Plane + memory v2 release — cycles C26–C29
|
|
1239
|
-
(TASK-001..011, CI-201..206, DOC-201..206, DOC-301..305).
|
|
1240
|
-
|
|
1241
|
-
- Memory v2: 4-type record model (`project_rule|derived_fact|episode|procedure`),
|
|
1242
|
-
atomic `records.json` store, backup-first idempotent migration from v1,
|
|
1243
|
-
`ukit memory v2` ops; legacy `.ukit/storage/memory/` layout and all
|
|
1244
|
-
`ukit memory *` subcommands stay compatible.
|
|
1245
|
-
- Code Intelligence Plane: `compileContext` → Context Packet v1 with
|
|
1246
|
-
peek→expand→read levels and token budgets; Task Router v1; freshness
|
|
1247
|
-
identity (`HEAD + overlay + config`) with incremental dirty-gated reuse
|
|
1248
|
-
and per-branch snapshots; hybrid retriever (exact + symbol + BM25 + RRF
|
|
1249
|
-
rerank); `impact.js` multi-hop impact sets; post-edit `invalidation.js`
|
|
1250
|
-
dirty-set and `diagnostics.js` feeding packet `next_actions`;
|
|
1251
|
-
`TypeScriptSemanticProvider` (optional `typescript` import — no new deps,
|
|
1252
|
-
degrade-safe).
|
|
1253
|
-
- CLI: `ukit code peek|search|context|impact` (`--depth`, `--diagnostics`),
|
|
1254
|
-
`ukit index status|doctor`, `ukit doctor --docs` (doc-contract checks).
|
|
1255
|
-
- Docs governance: `docContracts.js` validator (links/stale-version/
|
|
1256
|
-
render-vars/commands/budgets/registry/templates/dead-docs/archive-suggestions),
|
|
1257
|
-
`context_layers` + `docs=` route segment, host capability matrix, BUG docs
|
|
1258
|
-
consolidated into `docs/BUG_INDEX.md`, PROMPT_CACHING canonical/vendor split,
|
|
1259
|
-
`templates/instructions/core.md` 210→157 lines.
|
|
1260
|
-
- Benchmarks: `scripts/bench/runGold.mjs` scorecard + `--json` export.
|
|
1261
|
-
|
|
1262
|
-
## 2.6.7 - 2026-09-19
|
|
1263
|
-
|
|
1264
|
-
Canonical-instructions release — second roadmap slice (cycle C25,
|
|
1265
|
-
DOC-101..107). Root instruction files are now generated from a single
|
|
1266
|
-
canonical source; shipped contracts shrink ~25%; no runtime behavior
|
|
1267
|
-
changes to install/update/uninstall flows.
|
|
1268
|
-
|
|
1269
|
-
- **DOC-102 — deterministic core/overlay renderer**: new
|
|
1270
|
-
`templates/instructions/` sources (`core.md` + host overlays +
|
|
1271
|
-
`layout.yaml`) rendered by `src/render/instructionRenderer.js` into
|
|
1272
|
-
`templates/CLAUDE.md`, `templates/AGENTS.md`, `templates/.omp/RULES.md`
|
|
1273
|
-
(plus repo roots). `yarn docs:render` regenerates; `yarn docs:render:check`
|
|
1274
|
-
fails CI on drift. Byte-stable, duplicate-heading and unreferenced-section
|
|
1275
|
-
errors enforced.
|
|
1276
|
-
- **DOC-101 — rule IDs + markers**: `manifests/instructionRules.yaml`
|
|
1277
|
-
declares 48 rule IDs (kind/anchor/marker_in); `<!-- RULE: ID -->` markers
|
|
1278
|
-
embedded in the 4 marker-bearing artifacts; `instructionRules.test.js`
|
|
1279
|
-
validates schema, marker coverage, and anchor presence.
|
|
1280
|
-
- **DOC-103 — parity → semantic tests**: `contextDocsParity.test.js`
|
|
1281
|
-
rewritten to generated-artifact checks (drift, rule coverage, overlay
|
|
1282
|
-
heading allowlists, Release Policy boundary) + mutation tests that fail
|
|
1283
|
-
on dropped/duplicate critical rules; shared helpers in
|
|
1284
|
-
`tests/consistency/helpers/semanticParity.js`.
|
|
1285
|
-
- **DOC-104 — root contract shrink ≥25%**: `templates/CLAUDE.md` 288→214
|
|
1286
|
-
lines, `templates/AGENTS.md` 305→229 (all critical rule markers + anchors
|
|
1287
|
-
preserved). Internal-orchestration detail moved to new shipped doc
|
|
1288
|
-
`templates/docs/UKIT_INTERNALS.md` (manifest item `docs-ukit-internals`);
|
|
1289
|
-
baselines recorded in `docs/_baseline/root-instruction-lines.txt`.
|
|
1290
|
-
- **DOC-105 — WORKLOG rotation + MEMORY cleanup**: live worklog ≤500 lines,
|
|
1291
|
-
rotated entries preserved in `docs/WORKLOG_ARCHIVE.md` (regression-tested);
|
|
1292
|
-
MEMORY.md trimmed to durable protocol/decisions/constraints (≤200).
|
|
1293
|
-
- **DOC-106 — PROJECT_IMPORTANT_SPEC archived**: shrunk to a decision
|
|
1294
|
-
record; full body at `docs/archive/specs/PROJECT_IMPORTANT_SPEC.md`;
|
|
1295
|
-
`docArchiveLinks.test.js` guards link integrity.
|
|
1296
|
-
- **DOC-107 — user-first README**: rewritten around `ukit install` +
|
|
1297
|
-
natural-language flow (115 lines); maintainer detail moved to linked docs.
|
|
1298
|
-
|
|
1299
|
-
## 2.6.6 - 2026-09-19
|
|
1300
|
-
|
|
1301
|
-
Documentation-hardening release — first slice of the
|
|
1302
|
-
DOCUMENTATION_IMPROVEMENT_ROADMAP (cycle C24, DOC-001..006). No runtime
|
|
1303
|
-
behavior changes: new manifest, new test lanes, doc cleanup, and a shipped
|
|
1304
|
-
registry the remaining roadmap items build on.
|
|
1305
|
-
|
|
1306
|
-
- **DOC-001 — documentation ownership/source registry**: new
|
|
1307
|
-
`manifests/documentation.yaml` classifies every active/template/adapter
|
|
1308
|
-
documentation surface (owner, source_of_truth, merge_strategy, budget);
|
|
1309
|
-
`tests/consistency/docRegistry.test.js` enforces schema, coverage,
|
|
1310
|
-
manifest-skip consistency, and `error`-level budgets.
|
|
1311
|
-
- **DOC-002 — cross-agent benchmark corpus**: spec + fixtures at
|
|
1312
|
-
`tests/fixtures/doc-benchmark/` (corpus index, 4 host files, 9+ task files);
|
|
1313
|
-
`tests/benchmarks/docBenchmarkCorpus.test.js` validates structure and
|
|
1314
|
-
registry cross-references. Host execution deferred — every host recorded
|
|
1315
|
-
`untested` (roadmap-permitted).
|
|
1316
|
-
- **DOC-003 — artifact reference-integrity checks**:
|
|
1317
|
-
`tests/integration/artifactReferenceIntegrity.test.js` fails when a shipped
|
|
1318
|
-
instruction/skill doc references an internal command/file that does not
|
|
1319
|
-
resolve into the `npm pack` artifact (or the declared runtime-generated
|
|
1320
|
-
allowlist); wired into `yarn test:artifact`.
|
|
1321
|
-
- **DOC-004 — owner-file preservation lifecycle tests**:
|
|
1322
|
-
`tests/integration/ownerPreservation.test.js` proves install/update/uninstall
|
|
1323
|
-
preserve exact bytes and symlink posture of every registry
|
|
1324
|
-
`merge_strategy: skip` file, including BOM/CRLF and absent-seed cases.
|
|
1325
|
-
- **DOC-005 — STATUS.md live snapshot**: `docs/STATUS.md` shrunk to a ≤150-line
|
|
1326
|
-
live snapshot; historical body preserved verbatim at
|
|
1327
|
-
`docs/archive/status/STATUS-2026-09-19.md` with two-way links; registry
|
|
1328
|
-
budget for STATUS.md flipped to `error`.
|
|
1329
|
-
- **DOC-006 — adapter stale-version cleanup**: zero stale `vX.Y.Z` literals on
|
|
1330
|
-
adapter doc surfaces (`templates/.codex/README.md`, `.codex/README.md`,
|
|
1331
|
-
`templates/ukit/README.md`, `.ukit/README.md`); Codex-specific mechanics
|
|
1332
|
-
retained; `tests/consistency/staleVersionProse.test.js` enforces.
|
|
1333
|
-
|
|
1334
|
-
## 2.6.5 - 2026-09-19
|
|
1335
|
-
|
|
1336
|
-
Stability release — third full audit pass on the 2.6.4 base (cycle C23). 15
|
|
1337
|
-
confirmed hang / freeze / silent-idle / silent-hook defects fixed with root
|
|
1338
|
-
causes (full register: `docs/AI_HANDOFF/BUGS-2.6.5.md`). Headline: the
|
|
1339
|
-
cycle-lockout mechanism observed live — stall-breaker streak erased on every
|
|
1340
|
-
Stop (BUG-C23-08) — is fixed, plus every blocking/degrading hook exit now
|
|
1341
|
-
announces on stdout per the §8 contract.
|
|
1342
|
-
|
|
1343
|
-
- **Handoff stall-breaker streak erased every Stop (BUG-C23-08)**: the
|
|
1344
|
-
stop-coordinator dedupe write dropped the `handoff` streak slot, so a wedged
|
|
1345
|
-
RUN.md blocked forever; the merge now preserves the slot.
|
|
1346
|
-
- **Silent non-zero hook failures (BUG-C23-01, -03, -07)**: stderr-only crash
|
|
1347
|
-
reasons in `output-compression`/`reinject-context`/`post-edit-verify`/
|
|
1348
|
-
`stale-spec-check` + safe-patch exit-2 blocks + skill-router/vision-router/
|
|
1349
|
-
task-watchdog crash paths now emit a stdout `systemMessage` before exit —
|
|
1350
|
-
stderr reaches the model, not reliably the user.
|
|
1351
|
-
- **Lost/silent deadline emissions (BUG-C23-02, -04)**: compact-threshold's
|
|
1352
|
-
same-tick `stdout.write` + `process.exit(0)` discarded the announce on pipes
|
|
1353
|
-
(now `writeSync(1)`); deadline-expiry `exit 0` in four called `.mjs` modules
|
|
1354
|
-
and `handoff-resume.sh` now announces the abandoned work.
|
|
1355
|
-
- **Fail-open/crash-posture defects (BUG-C23-05, -06)**: context-hardcap-gate's
|
|
1356
|
-
internal-error catch now announces its fail-open degrade on stdout;
|
|
1357
|
-
completion-gate's stop-coordinator spawn is deadline-bounded with forced
|
|
1358
|
-
kill routed to the crash-streak path.
|
|
1359
|
-
- **Silent data loss + hot-path waste (BUG-C23-09, -10, -13, -15)**:
|
|
1360
|
-
compact-threshold double capacity read collapsed to one; sweep failures
|
|
1361
|
-
announce (bounded) instead of `.catch(() => {})` swallow; unguarded `main()`
|
|
1362
|
-
no longer runs on `import()`; route-audit merges stage to an append-only
|
|
1363
|
-
sidecar with a locked + exit fold instead of silently dropping entries.
|
|
1364
|
-
- **Unbounded growth + unbounded stalls (BUG-C23-11, -12, -14)**: memory
|
|
1365
|
-
project archives capped; `permission-audit.log` rotates inside the prune
|
|
1366
|
-
pass; `bin/ukit` runCli got an overall deadline watchdog.
|
|
1367
|
-
|
|
1368
|
-
Regression gate: 138 files / 2360 tests green, liveness 120/120, §8 sweep
|
|
1369
|
-
clean, template↔installed parity byte-identical, async-caller census clean,
|
|
1370
|
-
all 25 C22 verdicts re-confirmed.
|
|
1371
|
-
|
|
1372
|
-
## 2.6.4 - 2026-09-19
|
|
1373
|
-
|
|
1374
|
-
Stability release — second full audit pass on the 2.6.3 base (cycle C22). 25
|
|
1375
|
-
confirmed hang / freeze / silent-idle / silent-hook defects fixed with root causes
|
|
1376
|
-
(full register: `docs/AI_HANDOFF/BUGS-2.6.4.md`), including the two carried C21
|
|
1377
|
-
residual lineages: **R1** (sync fs inside the hardcap-gate deadline → BUG-C22-01)
|
|
1378
|
-
and **R2** (sync fs inside handoff-resume deadline → BUG-C22-02).
|
|
1379
|
-
|
|
1380
|
-
- **Stalled-mount freeze in the compact/hardcap chain (R1, BUG-C22-01)**: the entire
|
|
1381
|
-
compact-threshold fs path (`readNegotiatedCapacity`, `buildCompactThresholds`,
|
|
1382
|
-
`resolveCompactPhase`, `buildCompactPressureState`, pressure registration, compact
|
|
1383
|
-
plan) converted to async `fsp`; every caller in hooks, `src/core/compact`, and
|
|
1384
|
-
`src/core/status.js` now awaits; `isDirectRun` in `execution-ledger.mjs` /
|
|
1385
|
-
`stop-coordinator.mjs` guards `realpathSync` behind a basename check so module
|
|
1386
|
-
import no longer does sync fs on the deadline path.
|
|
1387
|
-
- **handoff-resume sync fs inside its deadline (R2, BUG-C22-02)** plus the same
|
|
1388
|
-
sync-fs-under-deadline class in `reset-compact-pressure` (BUG-C22-03),
|
|
1389
|
-
`auto-prune-bash` (BUG-C22-04) and `auto-allow-bash` (BUG-C22-06) — sync calls
|
|
1390
|
-
inside armed unref'd hook deadlines parked the event loop so the deadline itself
|
|
1391
|
-
could never fire.
|
|
1392
|
-
- **Stop-gate crash → block-forever (BUG-C22-05)**: a `completion-gate` coordinator
|
|
1393
|
-
crash used to produce an unconditional `decision:block` on every subsequent stop.
|
|
1394
|
-
New crash-streak breaker (`.ukit/storage/cache/completion-gate-crash.streak`,
|
|
1395
|
-
byte-append counter, cap `handoff.fullstack.stopGateMaxCrashStreaks` default 3)
|
|
1396
|
-
fails closed up to the cap then releases *loud*; omp bridge got the identical
|
|
1397
|
-
try/catch + streak parity.
|
|
1398
|
-
- **Silent `exit 0` / `exit(0)` clusters now announce (§8 contract)**: missing-script
|
|
1399
|
-
silent exits in 4 hooks (BUG-C22-07), deadline-expiry silent exits on 3 gated
|
|
1400
|
-
hooks plus the hardcap-gate sites (BUG-C22-08), and `verification-guard`
|
|
1401
|
-
stderr-only catch-all (BUG-C22-09) all emit the fail-open §8 JSON + stderr
|
|
1402
|
-
announcement instead of going quiet.
|
|
1403
|
-
- **Unbounded spawned children get self-deadlines**: `compact-threshold.mjs`
|
|
1404
|
-
`runCli` (BUG-C22-15, `UKIT_COMPACT_CLI_DEADLINE_MS` default 4s — bounds the
|
|
1405
|
-
skill-router spawn too), `verify-release.mjs` spawn watchdog + tree-kill
|
|
1406
|
-
(BUG-C22-20), `parallel-agents.mjs` per-level `UKIT_BENCH_LEVEL_TIMEOUT_MS`
|
|
1407
|
-
30 min deadline with detached process-group SIGKILL (BUG-C22-21).
|
|
1408
|
-
- **`withFileLock` busy-spin on persistent stat errors (BUG-C22-16 runtime twin /
|
|
1409
|
-
BUG-C22-17 core twin)**: a stat error inside the retry loop skipped both the
|
|
1410
|
-
`maxWaitMs` check and the backoff sleep → tight spin for the process lifetime.
|
|
1411
|
-
The budget check now lives inside the catch path in both copies.
|
|
1412
|
-
- **Sweep caps that could never fire (BUG-C22-10/11/12/13)**: tee sweep,
|
|
1413
|
-
hook-telemetry sweep, exec-ledger sweep and omp hook-errors sweep computed the
|
|
1414
|
-
overflow from the scanned *sample* instead of the eligible count, so configured
|
|
1415
|
-
max-file caps (e.g. `TEE_MAX_FILES=500`) were unreachable.
|
|
1416
|
-
- **Vision receipts eaten / orphaned (BUG-C22-14/18)**: the vision marker-dir sweep
|
|
1417
|
-
deleted `analyzed-*` receipts early → vision gate re-fired on every image; legacy
|
|
1418
|
-
top-level `vision/analyzed-*.json` orphans had no pruner at all. Both prunes are
|
|
1419
|
-
now TTL-coordinated with `isAlreadyAnalyzed`.
|
|
1420
|
-
- **Unbounded `.bak` backup writer (BUG-C22-19)**: `.claude/ukit/.ukit/backups/`
|
|
1421
|
-
had no pruner; `pruneOldBackups` is now bounded.
|
|
1422
|
-
- **Registered timeout < internal worst case (BUG-C22-22/23/24/25)**:
|
|
1423
|
-
`auto-allow-bash` 12→15s (internal worst 13s), `skill-router` 8→20s (worst
|
|
1424
|
-
~18s), `stale-spec-guard` 8→11s, and omp chain child budget 4000→12000ms with
|
|
1425
|
-
ceiling 16000→52000ms — a hook can no longer be killed mid-work by a budget
|
|
1426
|
-
smaller than its own declared worst case.
|
|
1427
|
-
|
|
1428
|
-
Full fixed-bug enumeration (register `docs/AI_HANDOFF/BUGS-2.6.4.md`, all rows
|
|
1429
|
-
`verified`):
|
|
1430
|
-
|
|
1431
|
-
- BUG-C22-01 (R1) — sync fs inside hardcap-gate deadline (compact-threshold + execution-ledger)
|
|
1432
|
-
- BUG-C22-02 (R2) — sync fs inside handoff-resume deadline
|
|
1433
|
-
- BUG-C22-03 — reset-compact-pressure: swallow-all + sync fs inside deadline
|
|
1434
|
-
- BUG-C22-04 — auto-prune-bash: invisible prune failure + sync fs inside deadline
|
|
1435
|
-
- BUG-C22-05 — completion-gate: persistent coordinator crash → block-forever (crash-streak breaker)
|
|
1436
|
-
- BUG-C22-06 — auto-allow-bash: sync fs inside deadline
|
|
1437
|
-
- BUG-C22-07 — missing-script silent `exit 0` cluster (4 hooks)
|
|
1438
|
-
- BUG-C22-08 — deadline-expiry silent `exit(0)` on gated hooks (+ hardcap sites)
|
|
1439
|
-
- BUG-C22-09 — verification-guard catch-all degrade on stderr only
|
|
1440
|
-
- BUG-C22-10 — tee sweep count cap unreachable (`TEE_MAX_FILES=500` never enforced)
|
|
1441
|
-
- BUG-C22-11 — hook-telemetry sweep count cap unreachable
|
|
1442
|
-
- BUG-C22-12 — exec-ledger sweep count cap unreachable
|
|
1443
|
-
- BUG-C22-13 — omp hook-errors sweep count cap unreachable
|
|
1444
|
-
- BUG-C22-14 — vision marker-dir sweep deletes `analyzed-*` receipts → vision gate re-fires
|
|
1445
|
-
- BUG-C22-15 — skill-router spawns `node compact-threshold.mjs prompt` unbounded (CLI self-deadline)
|
|
1446
|
-
- BUG-C22-16 — `withFileLock` busy-spin, runtime twin (token-utils.mjs)
|
|
1447
|
-
- BUG-C22-17 — `withFileLock` busy-spin, core twin (fileOps.js)
|
|
1448
|
-
- BUG-C22-18 — legacy top-level `vision/analyzed-*.json` orphans never pruned
|
|
1449
|
-
- BUG-C22-19 — `.claude/ukit/.ukit/backups/` `.bak` writer had no pruner
|
|
1450
|
-
- BUG-C22-20 — `verify-release.mjs` spawns with no timeout (watchdog + tree-kill)
|
|
1451
|
-
- BUG-C22-21 — `parallel-agents.mjs` N children, no timeout, partial-settle hang
|
|
1452
|
-
- BUG-C22-22 — auto-allow-bash registered timeout < internal worst case (12→15s)
|
|
1453
|
-
- BUG-C22-23 — skill-router registered timeout < internal worst case (8→20s)
|
|
1454
|
-
- BUG-C22-24 — stale-spec-guard registered timeout zero margin (8→11s)
|
|
1455
|
-
- BUG-C22-25 — omp chain child budget < every staged hook's worst case (4000→12000ms, ceiling 16000→52000ms)
|
|
1456
|
-
|
|
1457
|
-
## 2.6.3 - 2026-09-18
|
|
1458
|
-
|
|
1459
|
-
- **Informational side-questions no longer self-block at the Stop gate.** A no-route
|
|
1460
|
-
question ("còn task nào dang dở không", "is anything left?") carries the previous
|
|
1461
|
-
prompt's `routeSummary` + completion contract forward by design — but the prompt
|
|
1462
|
-
re-key also empties the effective ledger, so the gate unconditionally demanded
|
|
1463
|
-
write+verification evidence that a text answer can never produce. `evaluateCompletion`
|
|
1464
|
-
now releases silently when the turn is informational (routed `executionMode` or the
|
|
1465
|
-
no-route branch's `routingContext.executionMode` marker). The carried contract stays
|
|
1466
|
-
on disk, so real work after the question is still gated. One fix covers Claude Code
|
|
1467
|
-
and omp (shared `execution-ledger.mjs`). 3 regression tests.
|
|
1468
|
-
|
|
1469
|
-
## 2.6.2 - 2026-09-18
|
|
1470
|
-
|
|
1471
|
-
Stability release — 17 confirmed hang / silent-idle / silent-hook defects fixed with
|
|
1472
|
-
root causes (full register: `docs/AI_HANDOFF/BUGS-2.6.2.md`), focused on the "system
|
|
1473
|
-
goes quiet and does nothing" failure class.
|
|
1474
|
-
|
|
1475
|
-
- **No more silent degraded input (§8 contract, BUG-C21-01/02/03)**: every hook that
|
|
1476
|
-
stages stdin now *announces* when staging fails, stalls, or refuses — fail-closed
|
|
1477
|
-
hooks (`block-dangerous`, `protect-files`, `context-hardcap-gate`,
|
|
1478
|
-
`sensitive-data-guard`) emit an `ask`/`systemMessage` JSON + stderr `BLOCKED` +
|
|
1479
|
-
`exit 2`; advisory hooks emit a `systemMessage` + `exit 0`. The old bounded-cut
|
|
1480
|
-
silently truncated and proceeded. `sensitive-data-guard` now also fails closed
|
|
1481
|
-
(exit 2) on any non-refusal staging failure and mirrors its deadline block to
|
|
1482
|
-
stderr so the reason survives `process.exit` truncation.
|
|
1483
|
-
- **Sync-fs defeats inside self-deadlines removed (BUG-C21-04)**: the
|
|
1484
|
-
deadline-guarded node blocks in `context-window-guard.sh` and
|
|
1485
|
-
`context-hardcap-gate.sh` no longer call synchronous `fs.*Sync` inside the kill
|
|
1486
|
-
window — a stalled `.ukit` mount can no longer freeze the gate so the unref'd
|
|
1487
|
-
deadline can never fire. (Residual call-frame past the grant boundary recorded
|
|
1488
|
-
as BUG-C21-R1/R2 for the next cycle.)
|
|
1489
|
-
- **Host timeouts now cover internal budgets (BUG-C21-05)**: every hook registered
|
|
1490
|
-
in `settings.json` gets `timeout ≥ stage bound + node deadline + margin`; the
|
|
1491
|
-
regression test iterates *all* registered hooks instead of a hand-picked list,
|
|
1492
|
-
and four under-budgeted hooks (`auto-allow-bash`, `vision-router`,
|
|
1493
|
-
`auto-prune-bash`, `reset-compact-pressure`) were raised to 12s.
|
|
1494
|
-
- **Missing `await` in skill-router (review major)**: an un-awaited async context
|
|
1495
|
-
snapshot could leave the router writing a stale/partial decision silently.
|
|
1496
|
-
- **CLI hangs eliminated (BUG-C21-12/13)**: `ukit update`/`install` `spawnSync`
|
|
1497
|
-
calls now carry timeouts with `killSignal: 'SIGKILL'` (SIGTERM alone keeps
|
|
1498
|
-
waiting on signal-resistant children); a wedged `npm root -g` now reports
|
|
1499
|
-
`global-root-query-timeout` instead of hanging forever. `promptYesNo` resolves
|
|
1500
|
-
on stdin EOF / `ERR_USE_AFTER_CLOSE` instead of hanging or crashing.
|
|
1501
|
-
- **Unbounded growth bounded (BUG-C21-07/09/10/11)**: `tee/` cache, hook-errors
|
|
1502
|
-
jsonl, hook-latency and exec-ledger session files now have count caps; the
|
|
1503
|
-
ledger protects receipt files (`.journal`, `.quarantine`, gate crash counter)
|
|
1504
|
-
from count-cap eviction, and the overflow basis uses `entries.length`
|
|
1505
|
-
(post-filter) everywhere.
|
|
1506
|
-
- **Handoff machinery (BUG-C21-15)**: all four `RUN.md` `Phase:` consumers treat
|
|
1507
|
-
`done`/`blocked` as terminal — dead runs no longer nag forever.
|
|
1508
|
-
- **Wiring/parity (BUG-C21-06/08/14/16/17)**: stale-spec-guard deadline exits
|
|
1509
|
-
distinguish "stall" from "match"; `provision-worktree` spawnSync calls carry
|
|
1510
|
-
timeouts; `task-budget-validator.mjs` is manifest-covered so the documented
|
|
1511
|
-
command works on fresh installs; `verification-guard.sh` is actually registered
|
|
1512
|
-
on PreToolUse Bash; shipped `config.json` regained `codex-context-budget` and a
|
|
1513
|
-
parity test now guards every array field under `subagents`.
|
|
1514
|
-
|
|
1515
|
-
|
|
1516
|
-
## 2.6.0 - 2026-09-18
|
|
1517
|
-
|
|
1518
|
-
handoff-fullstack v2 — autonomous create→implement→review loop that runs to 100%
|
|
1519
|
-
completion of all unfinished work, with a mandatory detailed SPEC gate.
|
|
1520
|
-
|
|
1521
|
-
- **Never-stall Stop gate**: `stop-coordinator.mjs` gained a `handoff-cursor` lane that
|
|
1522
|
-
reads `docs/AI_HANDOFF/RUN.md` and blocks stop while `Phase:` is not `done`/`blocked`,
|
|
1523
|
-
carrying the cursor's `Next:` step. Liveness breaker releases after
|
|
1524
|
-
`handoff.fullstack.stopGateMaxStalledBlocks` (default 12) un-advancing blocks; the lane
|
|
1525
|
-
is advisory-on-failure and `stopGateEnabled` disables it. This fixes the observed
|
|
1526
|
-
stall where runs went idle after a recap.
|
|
1527
|
-
- **Mandatory SPEC.md**: `handoff-create` now writes `docs/AI_HANDOFF/SPEC.md` — a
|
|
1528
|
-
15-section template (FRs with Given/When/Then, fullstack scope, data model, API
|
|
1529
|
-
contract, test matrix, acceptance criteria, migration, chosen defaults) — before task
|
|
1530
|
-
files. Tasks carry `Spec references`; `handoff-model-guard.sh` hard-blocks fresh task
|
|
1531
|
-
creation without a filled SPEC (`handoff.fullstack.specRequired=false` bypasses).
|
|
1532
|
-
- **Autonomous loop contract**: only `HANDOFF FULLSTACK COMPLETE` / `HANDOFF FULLSTACK
|
|
1533
|
-
BLOCKED` may end a run; `CHECKPOINT — WORK CONTINUING` replaces bare recaps. Phase 0
|
|
1534
|
-
sweeps INDEX, task files, HISTORY/archive, `docs/TASKS.md` Ready-for-AI, and
|
|
1535
|
-
uncommitted WIP. Stuck tasks recover via `cancelled_superseded` + `TASK-xxx-R<n>`
|
|
1536
|
-
replacement tasks. `idleWatchdogMin` (default 5) scheduled wakeups keep the loop
|
|
1537
|
-
alive off-Claude-Code; `quietScansRequired` (default 2) guards completion; Phase F
|
|
1538
|
-
runs docs sync → `archive/cycle-NN/` → `Phase: done` → marker report.
|
|
1539
|
-
- **Agent + command wiring**: `handoff-planner` gained legacy-sweep and SPEC-authoring
|
|
1540
|
-
phases; `code-reviewer` reviews SPEC+PLAN against a 7-point spec quality gate;
|
|
1541
|
-
`feature-implementer` treats `Spec references` as outranking task-file wording;
|
|
1542
|
-
`handoff-implement`/`handoff-review` codify spec-is-the-contract and `-R<n>`
|
|
1543
|
-
recovery. `handoff-clear` archives SPEC and must close RUN.md; `handoff-status`
|
|
1544
|
-
reports run phase + spec presence.
|
|
1545
|
-
- **Config**: `handoff.fullstack.*` keys (`stopGateEnabled`, `stopGateMaxStalledBlocks`,
|
|
1546
|
-
`quietScansRequired`, `idleWatchdogMin`, `autoArchive`, `specRequired`) with `_help`
|
|
1547
|
-
notes, in both config trees.
|
|
1548
|
-
|
|
1549
|
-
## 2.5.1 - 2026-09-18
|
|
1550
|
-
|
|
1551
|
-
Post-release fixes from C20 reviewer advisories.
|
|
1552
|
-
|
|
1553
|
-
- **`projectImportant` symlink fallback**: the odd-open-error fallback imported
|
|
1554
|
-
`stat` as `lstat`, which follows symlinks — a broken symlink degraded to `missing`
|
|
1555
|
-
and a symlink-to-file to `unreadable` instead of the correct `unsafe-type`
|
|
1556
|
-
(spec §6.1). Now uses real `lstat`; fixed in `src/core/projectImportant.js` and both
|
|
1557
|
-
installed-runtime mirrors (parity-locked).
|
|
1558
|
-
- **`inspectProjectImportant` guard**: a non-string `projectRoot` threw `TypeError`
|
|
1559
|
-
before the try block instead of degrading to `unreadable`, violating the
|
|
1560
|
-
"no documented entry point throws" contract. Guard added (src + mirrors).
|
|
1561
|
-
- **Uninstall backup exhaustion surfaced**: when all 101 backup-name candidates are
|
|
1562
|
-
occupied, `PROJECT_IMPORTANT.md` was silently preserved without a backup. Now emits
|
|
1563
|
-
a `backupWarnings` line telling the user to free a slot and rerun.
|
|
1564
|
-
- **Test tightening**: the packed-install smoke now asserts `project-important.sh`
|
|
1565
|
-
occupies SessionStart index 0 (spec: must run first); corrected a stale "~5%
|
|
1566
|
-
headroom" comment on the unpacked-size ceiling.
|
|
1567
|
-
|
|
1568
1198
|
## 2.5.0 - 2026-09-18
|
|
1569
1199
|
|
|
1570
1200
|
C20 feature release (14 handoff tasks + 3-lane review): project-owner instructions via a
|
|
@@ -1627,4 +1257,4 @@ UKit gains a fourth runtime and loses one. omp (Oh My Pi) reaches full parity wi
|
|
|
1627
1257
|
|
|
1628
1258
|
**Known gap:** omp was not installed on the machine this release was built on. Every omp assertion in the test suite is against the generated `.omp/` tree and the bridge module's shape — the packed-tarball test confirms `templates/.omp/**` actually publishes, but nothing here observes omp's runtime behaviour. The three open verification items (does omp emit `session_start` on a post-compact continuation; omp's real base-URL key; omp's real auto-compact-window key) are recorded in `docs/AI_HANDOFF/tasks/TASK-008.md`.
|
|
1629
1259
|
|
|
1630
|
-
Entries for 3.0.0 through 3.3.3 are archived in the repository at `docs/archive/CHANGELOG-3.0.0-3.3.3.md`. Entries for 2.1.0 through 2.2.16 (except 2.2.1) are archived in the repository at `docs/archive/CHANGELOG-2.1.0-2.2.16.md`; entries for 1.0.0 through 2.0.7 at `docs/archive/CHANGELOG-1.0.0-2.0.7.md` (neither is shipped in the npm package).
|
|
1260
|
+
Entries for 2.5.1 through 2.6.11 are archived in the repository at `docs/archive/CHANGELOG-2.5.1-2.6.11.md`. Entries for 3.0.0 through 3.3.3 are archived in the repository at `docs/archive/CHANGELOG-3.0.0-3.3.3.md`. Entries for 2.1.0 through 2.2.16 (except 2.2.1) are archived in the repository at `docs/archive/CHANGELOG-2.1.0-2.2.16.md`; entries for 1.0.0 through 2.0.7 at `docs/archive/CHANGELOG-1.0.0-2.0.7.md` (neither is shipped in the npm package).
|
package/package.json
CHANGED
|
@@ -60,7 +60,7 @@ function failureCommands(ledger) {
|
|
|
60
60
|
const verifyFail = receipt.success === false && VERIFY_KINDS.has(receipt.kind);
|
|
61
61
|
const ledgerFlagged = ledger.verificationFailed === true && receipt.success === false;
|
|
62
62
|
if (verifyFail || ledgerFlagged) {
|
|
63
|
-
out.push({ command: receipt.command, file: receipt.file });
|
|
63
|
+
out.push({ command: receipt.command, file: receipt.file, exitCode: receipt.exitCode });
|
|
64
64
|
}
|
|
65
65
|
}
|
|
66
66
|
return out;
|
|
@@ -87,7 +87,7 @@ async function writeArtifact(filePath, result) {
|
|
|
87
87
|
* @returns {Promise<{generatedAt: string, ledgersScanned: number,
|
|
88
88
|
* ledgersWithFailures: number,
|
|
89
89
|
* patterns: Array<{signature: string, count: number, sessions: number,
|
|
90
|
-
* commands: string[], files: string[]}>}>} never throws.
|
|
90
|
+
* commands: string[], files: string[], exitCodes: number[]}>}>} never throws.
|
|
91
91
|
*/
|
|
92
92
|
export async function mineFailurePatterns(projectRoot, { limitLedgers = 500, minCount = 2 } = {}) {
|
|
93
93
|
const result = {
|
|
@@ -116,15 +116,17 @@ export async function mineFailurePatterns(projectRoot, { limitLedgers = 500, min
|
|
|
116
116
|
result.ledgersWithFailures += 1;
|
|
117
117
|
const sessionId = typeof ledger.sessionId === 'string' ? ledger.sessionId : null;
|
|
118
118
|
|
|
119
|
-
for (const { command, file } of failures) {
|
|
119
|
+
for (const { command, file, exitCode } of failures) {
|
|
120
120
|
const signature = normalizeSignature(command);
|
|
121
121
|
if (!signature) continue;
|
|
122
122
|
let group = groups.get(signature);
|
|
123
123
|
if (!group) {
|
|
124
|
-
group = { signature, count: 0, sessions: new Set(), commands: [], files: [] };
|
|
124
|
+
group = { signature, count: 0, sessions: new Set(), commands: [], files: [], exitCodes: new Set() };
|
|
125
125
|
groups.set(signature, group);
|
|
126
126
|
}
|
|
127
127
|
group.count += 1;
|
|
128
|
+
// Real failure evidence: non-zero exit status the host reported (e.g. 124 = timeout(1)).
|
|
129
|
+
if (Number.isInteger(exitCode) && exitCode !== 0) group.exitCodes.add(exitCode);
|
|
128
130
|
if (sessionId) group.sessions.add(sessionId);
|
|
129
131
|
if (group.commands.length < MAX_EXAMPLES && !group.commands.includes(command)) {
|
|
130
132
|
group.commands.push(command);
|
|
@@ -145,6 +147,7 @@ export async function mineFailurePatterns(projectRoot, { limitLedgers = 500, min
|
|
|
145
147
|
sessions: g.sessions.size,
|
|
146
148
|
commands: g.commands,
|
|
147
149
|
files: g.files,
|
|
150
|
+
exitCodes: [...g.exitCodes].sort((a, b) => a - b),
|
|
148
151
|
}));
|
|
149
152
|
} catch {
|
|
150
153
|
// Any unexpected failure → return whatever was computed so far.
|
package/src/index/taskRouting.js
CHANGED
|
@@ -1205,8 +1205,16 @@ export function classifyAdvisorRequest({ promptText = '', commandText = '' } = {
|
|
|
1205
1205
|
// Used by hasAdvisoryOnlyAssertion (assertion + no order → advisory) and by
|
|
1206
1206
|
// the question-shape gate below (question + no order → consult, even when the
|
|
1207
1207
|
// subject vocabulary fires debug/failure/implement signal scores).
|
|
1208
|
+
// Tone-folding turns unrelated Vietnamese words into mutation verbs: "cái" (this/thing) -> cai (cài),
|
|
1209
|
+
// "chính" (main/exactly) -> chinh (chỉnh), "Việt" -> viet (viết), "đợi/đời/đối/đội/đôi/đói" -> doi (đổi),
|
|
1210
|
+
// "cháy/chảy" -> chay (chạy), "sữa" -> sua (sửa), "thèm" -> them (thêm). Mask the toned non-verbs BEFORE
|
|
1211
|
+
// folding, so "Cái này là gì, tại sao…" is not read as an install order (3.4.81 telemetry audit).
|
|
1212
|
+
function maskTonedNonVerbs(text = '') {
|
|
1213
|
+
return String(text ?? '').normalize('NFC')
|
|
1214
|
+
.replace(/(?<![\p{L}\p{N}_])(?:cái|chính|việt|cháy|chảy|đợi|đời|đối|đội|đôi|đói|sữa|thèm)(?![\p{L}\p{N}_])/giu, ' _ ');
|
|
1215
|
+
}
|
|
1208
1216
|
function hasMutationOrder({ promptText = '', commandText = '' } = {}) {
|
|
1209
|
-
const folded = `${promptText
|
|
1217
|
+
const folded = `${maskTonedNonVerbs(promptText)}\n${commandText ?? ''}`
|
|
1210
1218
|
.toLowerCase()
|
|
1211
1219
|
.normalize('NFD')
|
|
1212
1220
|
.replace(/[\u0300-\u036f]/g, '')
|
|
@@ -1255,6 +1263,11 @@ function hasMutationOrder({ promptText = '', commandText = '' } = {}) {
|
|
|
1255
1263
|
.replace(new RegExp(`\\b(?:neu|if)\\s+${TAIL}`, 'g'), ' ')
|
|
1256
1264
|
// 'đối chiếu' (cross-check) folds to 'doi chieu' and is not the verb 'đổi' (change).
|
|
1257
1265
|
.replace(/\bdoi\s+chieu\b/g, ' ')
|
|
1266
|
+
// Field telemetry (3.4.81 audit): words that fold onto a mutation verb but name no order.
|
|
1267
|
+
// "chưa được cấp quyền cập nhật" / "no permission to update" reports a missing permission (read-only
|
|
1268
|
+
// banners); "gì thêm [nữa]" / "thêm nữa không" mean "anything else", never the verb "add".
|
|
1269
|
+
.replace(/\b(?:(?:chua|khong)\s+(?:duoc\s+)?(?:cap\s+|co\s+)?quyen|(?:no|not\s+have|without|lacks?)\s+permission\s+to)\s+(?:implement|apply|update|modify|add|create|ship|deliver|fix|refactor|remove|delete|rename|change|write|build|make|install|run|deploy|execute|edit|sua|them|tao|xoa|doi|thay\s+the|cap\s+nhat|viet|chay|cai|chinh|trien\s+khai|cau\s+hinh)(?![a-z0-9_])/g, ' ')
|
|
1270
|
+
.replace(/(?<=\b(?:con|co)\b[^.?!,;:\n]{0,40})\b(?:gi|nao)\s+them(?:\s+nua)?(?:\s+(?:khong|ko))?\s*(?=[?.!,;:\n]|$)|\bthem\s+nua\s+(?:khong|ko)\s*(?=[?.!\n]|$)/g, ' ')
|
|
1258
1271
|
// R2: a deferred/future clause ("chua sua", "how we would later implement", "R2 onward: ...",
|
|
1259
1272
|
// "fix strategy ...") describes work for a LATER round, not an order for this turn.
|
|
1260
1273
|
.replace(new RegExp(`\\bchua\\s+(?:sua|fix|implement|trien\\s+khai|viet|thay\\s+doi)\\b${WAIT_TAIL}`, 'g'), ' ')
|
|
@@ -1345,6 +1358,11 @@ function deriveExecutionMode({
|
|
|
1345
1358
|
const questionShapedPrompt = /\?\s*$/.test(trimmedPromptText)
|
|
1346
1359
|
|| leadingInterrogative
|
|
1347
1360
|
|| /\b(?:la gi|duoc khong)\s*$/.test(foldedPromptText)
|
|
1361
|
+
// Field telemetry (3.4.81 audit): a what-is question mid-prompt ("… là gì, …") and a leading
|
|
1362
|
+
// explanation request ("giải thích …", "explain …") ask for an answer. Both still pass the
|
|
1363
|
+
// hasMutationOrder/targetFile veto below, so "giải thích rồi sửa luôn" stays an order.
|
|
1364
|
+
|| /\bla gi\s*(?:[,.?!;:\n]|$)/.test(foldedPromptText)
|
|
1365
|
+
|| /^(?:hay\s+|vui\s+long\s+|lam\s+on\s+|please\s+)?(?:giai\s+thich|explain)\b/.test(foldedPromptText)
|
|
1348
1366
|
// C85-019: Vietnamese status/consult tails without a '?' — "… chưa",
|
|
1349
1367
|
// "… chưa nhỉ", "… không nhỉ", "… đúng không". A question about a defect
|
|
1350
1368
|
// must not mint the bug-fix floor's write/repro debt.
|