@ngockhoale/ukit 2.6.11 → 2.7.0

package/CHANGELOG.md

@@ -2,6 +2,76 @@
 
 All notable changes to UKit are documented here.
 
+## 2.7.0 - 2026-09-20
+
+OMP fully autonomous / unattended mode — cycle C35 (TASK-008..014).
+
+- Product knob: `orchestration.permissionMode: "unattended"` declared in
+  `.ukit/storage/config.json` (template + validation; `interactive`/`safe-auto`
+  reserved for later cycles).
+- omp approval surface: `templates/.omp/config.yml` pins
+  `tools.approvalMode: yolo` + all ten `tools.approval.*: allow` — the C34
+  key-omission decision is superseded now that unattended is a declared
+  product mode (interactive global configs can no longer downgrade shipped
+  installs to prompting).
+- Migration: `mergeOmpConfig` + diffPlan wiring — installed `.omp/config.yml`
+  upgrades preserve user `modelRoles`/keys/`bash.patterns`, strip legacy
+  `prompt` policies, stay idempotent, and never clobber unparseable files.
+- `ukit doctor`: new unattended section verifies `permissionMode`, the
+  rendered approval map, zero residual `prompt` policies, and the deny set.
+- Unattended completion loop: `## Unattended Completion Loop` guidance in
+  core.md (rendered to CLAUDE.md/AGENTS.md) — turns end only on completion
+  evidence, a real external blocker, or a deny; denies route to the next
+  safer alternative instead of a question.
+- `block-dangerous.sh`: name-allowlisted `rm` targets now require realpath
+  containment inside the project root (symlink-out / `..` traversal closed;
+  fail closed without realpath).
+- Residuals recorded in `docs/AI_REPORT/AI_FREEZE_AUDIT.md` §8: eval stays
+  `allow` ("dangerously autonomous, user-chosen" — eval-sandbox roadmap
+  note); opencode `bash."*": "ask"` flip deferred pending deny-set
+  expansion.
+
+## 2.6.11 - 2026-09-20
+
+Self-improvement roadmap phases 2–4 — cycle C32 (TASK-227..232).
+
+- Feedback capture: `src/diagnostics/feedbackEvents.js` + shared
+  `ledgerFiles.js` ledger helper record labeled wrong-route /
+  manual-correction events; new `ukit feedback` CLI.
+- Telemetry emitters: `skillIds` in route-audit entries
+  (`route-task.mjs`); retriever lane stats appended to
+  `.ukit/storage/cache/retriever-lanes.jsonl` (`retriever.js`).
+- Metrics surfaces: `skillAccuracy.js` + `laneStats.js` collectors;
+  `ukit metrics` gains `feedback`, `skills`, `retriever-lanes`, and
+  `learning` sections (lazy-import, degrade to `n/a`).
+- Memory promotion lanes: `ukit memory learn` mines failure-patterns into
+  pending `ukit memory propose` candidates (`src/learning/patternProposals.js`);
+  `ukit memory promote` writes approved project_rule/procedure records to
+  MEMORY.md `## ukit-learned`; `ukit memory episode` writes a session-episode
+  record (`--session`/`UKIT_SESSION_ID`/latest-ledger resolution) plus
+  SessionEnd hook `session-episode.sh`.
+- Tuning loop: `learning.*` config namespace (validated in runtimeConfig,
+  template parity) + `src/learning/tuning.js` advisory lane-weight/threshold
+  suggestions.
+- Verification: per-task suites green on rerun; `.claude`↔`templates/.claude`
+  parity diff-clean. 6/6 tasks approved, 0 fix rounds.
+
+## 2.6.10 - 2026-09-20
+
+Self-improvement roadmap phase 1 — cycle C31 (HK-401, SI-101..103).
+
+- Hook contract fix: centralized `ukit_emit_permission_decision` in
+  hook-input.sh (keyed on `UKIT_HOOK_CHAIN_RUNNER`); bypassPermissions probe
+  replaced with locked fallback deny + exit 0 — dead `ask` path closed.
+- Router v2-memory lane: `listMemoryItems` reads memory records.json;
+  `record` branch added to buildMemorySegments.
+- Outcome join: `routeOutcomes.js` joins route-audit × exec-ledger on
+  requestKey; new `ukit metrics` command surfaces route accuracy.
+- Failure mining: `failurePatterns.js` mines exec-ledger into
+  failure-patterns.json (advisory, never-throws).
+- Self-improvement surface audit + roadmap spec:
+  `docs/research/2026-09-20_self-improvement-surface-audit.md`.
+
 ## 2.6.9 - 2026-09-19
 
 Code Intelligence scale release — cycle C30 (CI-301..305, TASK-218..222).

package/manifests/documentation.yaml

@@ -319,6 +319,17 @@ entries:
     validation: [manual]
     archive_policy: never
     notes: handoff run cursor index; declared in context_layers.intents.handoff (DOC-201)
+  - id: docs-ai-report
+    path: docs/AI_REPORT/
+    class: runtime
+    audience: [agent, maintainer]
+    owner: runtime
+    source_of_truth: docs/AI_REPORT/
+    merge_strategy: none
+    load_policy: task-routed
+    validation: [manual]
+    archive_policy: never
+    notes: dir entry — covers all files beneath (handoff lane reports, perf findings)
   - id: docs-archive
     path: docs/archive/
     class: archive
@@ -405,8 +416,8 @@ entries:
     load_policy: task-routed
     validation: [manual]
     archive_policy: never
-    budget: { max_lines: 160, enforcement: warning }
-    notes: 'canonical shared core; TASK-217 shrink 210→157 lines — budget 160 kept per SPEC §14 (now obeyed, warning-only)'
+    budget: { max_lines: 172, enforcement: warning }
+    notes: 'canonical shared core; TASK-217 shrink 210→157 lines, TASK-004 Stall & Wait Reporting block →165 lines — budget raised 160→172 keeping ~4% headroom (warning-only)'
   - id: tpl-omp-rules
     path: templates/.omp/RULES.md
     class: generated
@@ -531,6 +542,17 @@ entries:
     validation: [manual]
     archive_policy: never
     notes: dir sourceTemplate docs/AI_HANDOFF → dir path ending '/'
+  - id: docs-ukit-internals
+    path: docs/UKIT_INTERNALS.md
+    class: canonical
+    audience: [agent, maintainer]
+    owner: product
+    source_of_truth: docs/UKIT_INTERNALS.md
+    merge_strategy: none
+    load_policy: task-routed
+    validation: [manual]
+    archive_policy: never
+    notes: installed copy of templates/docs/UKIT_INTERNALS.md (merge_strategy overwrite_with_backup on install); internal-orchestration detail extracted from core.md (DOC-104)
   - id: tpl-docs-ukit-internals
     path: templates/docs/UKIT_INTERNALS.md
     class: canonical

package/manifests/instructionRules.yaml

@@ -87,6 +87,68 @@ rules:
       `WAITING ON YOU:` line and schedule a one-shot wakeup when available.
     marker_in: [templates/CLAUDE.md, templates/AGENTS.md, CLAUDE.md, AGENTS.md]
     anchor: 'WAITING ON YOU'
+  - id: STALL-01
+    title: Announce waits, verbatim errors, never blame UKit
+    kind: critical
+    statement: >-
+      Long waits must be announced as they happen, any API/tool error surfaced
+      verbatim in the same turn, and network/API named as the suspect on long
+      stalls — never blaming UKit.
+    marker_in: [templates/CLAUDE.md, templates/AGENTS.md, CLAUDE.md, AGENTS.md]
+    anchor: 'never blame UKit'
+  - id: STALL-02
+    title: Stall rules extend every-stop-says-why
+    kind: supporting
+    statement: >-
+      The stall/wait reporting rules extend the existing every-stop-names-its-
+      reason contract (EXEC-04); they do not replace it.
+    marker_in: [templates/CLAUDE.md, templates/AGENTS.md, CLAUDE.md, AGENTS.md]
+  - id: STALL-03
+    title: One-tool-call workaround is conditional only
+    kind: critical
+    statement: >-
+      Falling back to one tool call per assistant message is a conditional
+      workaround that fires only when malformed/concatenated tool-input errors
+      are observed or the harness version is known-affected — healthy hosts
+      keep using legitimate parallel calls.
+    marker_in: [templates/CLAUDE.md, templates/AGENTS.md, CLAUDE.md, AGENTS.md]
+    anchor: 'one tool call per assistant message'
+  - id: UNATTENDED-01
+    title: Unattended mode runs the completion loop without a human
+    kind: critical
+    statement: >-
+      Under orchestration.permissionMode: unattended there is no human in the
+      loop — the loop is UNDERSTAND → PLAN → IMPLEMENT → VERIFY → FIX →
+      REVERIFY → COMPLETE, and a turn ends only on COMPLETE evidence, a real
+      external blocker, or a deny decision.
+    marker_in: [templates/CLAUDE.md, templates/AGENTS.md, CLAUDE.md, AGENTS.md]
+    anchor: 'no human in the loop'
+  - id: UNATTENDED-02
+    title: Bounded attempts and recovery strategies in unattended mode
+    kind: supporting
+    statement: >-
+      maxAttemptsPerFailure: 5 per failing verification and
+      maxRecoveryStrategies: 3 distinct fix strategies before reassessing —
+      do not loop one strategy.
+    marker_in: [templates/CLAUDE.md, templates/AGENTS.md, CLAUDE.md, AGENTS.md]
+  - id: UNATTENDED-03
+    title: BLOCKED only on true external blockers
+    kind: critical
+    statement: >-
+      Test failures, lint errors, review findings, and ambiguity are NOT
+      blockers — they feed FIX; BLOCKED is reserved for missing credentials,
+      dead services, and unreachable network/API.
+    marker_in: [templates/CLAUDE.md, templates/AGENTS.md, CLAUDE.md, AGENTS.md]
+    anchor: 'NOT blockers'
+  - id: UNATTENDED-04
+    title: A deny becomes the next safer alternative, never a question
+    kind: critical
+    statement: >-
+      A deny (hook/gate/permission) never becomes a question — pick the next
+      safer alternative that satisfies the intent and record the substitution;
+      never emit "do you want me to…".
+    marker_in: [templates/CLAUDE.md, templates/AGENTS.md, CLAUDE.md, AGENTS.md]
+    anchor: 'next safer alternative'
   - id: LONG-01
     title: LAND one thing, DEFER the rest, DELEGATE broad work
     kind: supporting

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ngockhoale/ukit",
-  "version": "2.6.11",
+  "version": "2.7.0",
   "description": "Install/update an index-first AI workspace for Claude Code, OpenAI Codex, OpenCode, and omp (Oh My Pi).",
   "license": "MIT",
   "type": "module",