diff --git a/.claude/memory/evals.md b/.claude/memory/evals.md index e60b934..bc17883 100644 --- a/.claude/memory/evals.md +++ b/.claude/memory/evals.md @@ -454,3 +454,10 @@ Dogfood: 3 blind lenses attacked the v1 plan for the plan-challenge feature itse - **Result**: prompt precision 57 % (n=75), under the 64 % (14/22) 4b-4d proof target. By trigger: fresh 78 % (n=23), idle 44 % (n=32), keyword 55 % (n=20). 3rd pick 35 % (n=23). Timeout + paused 0 %. Rerank p90 3355 ms. Tool gate 0/10 records. Misses: 7 on 5 asks, 4 no-overlap (BM25 cannot reach, title+lesson share no word with the prompt), 3 overlap (rerank dropped them). 1 noisy entry + 1 promotion candidate in the project repo, to run there via /prune-memory. - **Anomalies**: idle trigger fires on short follow-up prompts ("ok on y va") where all 3 entries miss; tool channel logged 0 records in 4 days, unproven live; rerank p90 above the 3 s bound. - **Action**: 4 TODO lines under Phase 4b-4d (idle trigger, cap at 2, tool channel check, no-overlap misses). Decisions deferred to the user. + +## EVAL-059 — BDR-125 close-out run: gates caught 4 misses, 2 doc drifts, my own 3 slips; all phases closed [keep] (2026-11-14) +- **Subject**: BDR-125 close-out run 2026-11-12/14, blocks A-J chained (user brief: questions once, ask merge prompt = signal). +- **Method**: per block one skill run + gitflow branch; GATE 0 gates.sh, fresh verifier, fresh security-auditor, challenge rounds on /bugfix and /feat plans; counted gate verdicts and re-dispatches from this session's reports. +- **Result**: 3.0.0 cut; E1/E2/G/H1-H7/BLK-049/Fable 5.2/perl runner merged, all `make test` green; F faunosteo branch ready (user merges). Every phase 0/0b/4a/1/5/2/4b-4d TODO line closed, BUGS-FOUND BDR-125 items fixed or closed with reason. +- **Anomalies**: (1) gates caught what executors missed: H6b verifier ECARTS (profile.test.sh:123 comment silenced shellcheck from that line on; my brief said "counts equal to HEAD", gate said "clean"); perl plan r1 FATAL(3) (impossible digest assertion without tty, single quotes = SC2016, timeout(1) misread) and r2 still let a fast-failing runner pass timing-only tests. (2) my own slips: contract without EVIDENCE lines (H7, gates ERROR), reconcile loop misused `reconcile_blk_open`, awk print `" || "` tripped our own H4a deny row. (3) E3 npm -g edit refused by auto-mode classifier twice → user hand edit. (4) one feater died on API 403 mid-run, gates proved its diff. (5) doc audits found real drift twice (CLAUDE.global.md `open: unknown` clause, MIGRATION step 1 restoring enabledPlugins only). +- **Action**: keep fresh gates + confirmation challenger (they paid every time); briefs quote the contract criterion verbatim, never paraphrase it; LRN-289/290 written; gates.sh label guard TODO stays. diff --git a/.claude/memory/journal.md b/.claude/memory/journal.md index 296c21e..b327861 100644 --- a/.claude/memory/journal.md +++ b/.claude/memory/journal.md @@ -753,3 +753,4 @@ rules: ## 2026-11-14 - /hotfix doctor-skilloverrides-residue (block H7): doctor residue check compares {enabledPlugins, skillOverrides} HEAD vs work (9c9a024); doc audit caught MIGRATION step 1 restoring enabledPlugins only → extended to both keys (8dcee03). Security PASS, 1 LOW (echo -e escapes) in BUGS-FOUND. H + I TODO lines closed; LRN-270 bisect files moved to plans/2026-11-14-lrn270-bisect/ for the user. - /bugfix perl-timeout-group: perl fallback runner leads own process group, kills group at deadline, TERMs leftovers (490ad83); challenge r1 FATAL (digest assertion impossible without tty, single quotes = SC2016, timeout(1) misread) → r3; verifier CONFORME (flip red), security PASS. Phase 4b-4d closed: digest-after-compaction proven by this run's auto-compact. Residue TODO: design-tool-gate.sh:70 same idiom. +- BDR-125 close-out J: /reconcile clean (BLK-017 + 3 upstream), CHANGELOG BDR-125 delivered + gate 1 57 % + H6a + BLK-049 (69eb53f), D4 BUGS-FOUND items closed with reason, block I + phase 4b-4d TODO closed, EVAL-059 capitalized; close ritual: nothing new (LRN-289/290 already). User leftovers: E3 npm -g hand edit, faunosteo merge, LRN-270 bisect, W4-B AC9, game status words, cc oom proof. diff --git a/.claude/tasks/TODO.md b/.claude/tasks/TODO.md index 5283fd2..53af8d3 100644 --- a/.claude/tasks/TODO.md +++ b/.claude/tasks/TODO.md @@ -15,7 +15,7 @@ Decisions at start (asked once): faunosteo memory v2 on `chore/memory-v2` off de - [x] G (2026-11-13): BLK-044 resolved, origin = sshd drop-in OOMScoreAdjust=-1000 inherited; lib/cc.sh raises oom_score_adj to 500 raise-only (6c10d3a, T7-oom, LRN-285); live proof `/proc//oom_score_adj` = 500 needs a fresh `cc` session (user, this session predates the fix) - [x] H (2026-11-13/14): H1 project-rules 9d5530a, H2 memory.sh d8e62c1 (BLK-038 resolved), H3 pack-census 2fb9a2f, H4a deny rows a43ba1b, H4b safety LOWs 1f107c3, H5 feedback 9f9412a (not done, justified: tsc = no compiler, `l:` 1.2 % of cap, image prompts 0 instances), H6a prose 77041e5, H6b scripts 9045a60, H7 doctor skillOverrides 9c9a024. BLK-049 strict types 08308a6; Fable 5.2 ca1dc9f. LRN-270 bisect handed to user (files in plans/2026-11-14-lrn270-bisect/). New pre-existing LOWs in BUGS-FOUND (not BDR-125). - [x] I (2026-11-14, from 240 subagent meta files): pins hold (sonnet executors, fable verifier/security 5.1→5.2, opus analyzer); parentModel = not in logs (meta has model + pickedEffortModel only), no unpinned spawn in a demoted turn → not reproducible from logs, reopen when one runs; W4-B AC9 = user step (`/route verbose on`, classifier line is UI-only); Fable-out-of-credits + StopFailure closed "not reproducible, reopen on incident". -- [ ] J /reconcile, CHANGELOG (gate 1 at 57 % delivered + each block), 0 BDR-125 line left, EVAL of the run, /close +- [x] J (2026-11-14): /reconcile (blocked BLK-017 + 3 upstream, tree clean, nothing unpushed), CHANGELOG BDR-125 delivered + gate 1 57 % + H6a + BLK-049 lines, 0 phase line open in TODO, BUGS-FOUND BDR-125 items fixed or closed with reason, EVAL-059, /close. ### Context mod queue (BDR-125): one /ship-feature per phase, order (user 2026-10-25): 0 → 4a → 1 → 5 → 2 → 3 (conditional) → 4b-4d Re-challenged 2026-10-25 on Fable (4 lenses + native-features guide, synthesis in `.claude/tasks/plans/2026-10-25-context-mod-queue-challenge.md` § RE-CHALLENGE; user picks: native ask merge gate, phase 3 conditional, rest in block). The plan halves: native carriers and existing files first, a mod only where a measured gain has no native lever. @@ -111,16 +111,16 @@ Scope (user 2026-11-11) [gated]: D1-D6 + the 6 dead deny entries + MIGRATION/CHA ### User decisions and hand edits - [ ] user decision: pin a commit for higgsfield-ai/skills and a version for `@higgsfield/cli` (security gate, 2 MEDIUM, accepted as is) -- [ ] user hand edit pending: remove the four `Bash(npm … -g|--global *)` lines from `permissions.deny` (they override the classifier); then the first global install is the live test of the entry +- [ ] user hand edit (E3, classifier refused the sub-agent edit 2026-11-13): on branch bugfix/npm-global-deny (exists, empty) remove from settings.json `permissions.deny` `Bash(npm install -g *)`, `Bash(npm i -g *)`, `Bash(npm install --global *)`, `Bash(npm i --global *)`; add the 4 rows to the push-guard removed-ok heredoc (lib/tests/push-guard.test.sh) under `# soft_deny "Global npm installs" decides (user 2026-11-12)` and set `removed_ok_until=3.1.0`; live test `npm install -g --dry-run cowsay`. ### Blocked - [x] launch claude with `TMPDIR=$HOME/.cache/claude-tmp` (BLK-021 resolved by `lib/cc.sh`, phase 5; `.bashrc` swap done, ~/.bashrc:133) ### Model-router live checks (opportunistic, need a real session or incident) -- [ ] LIVE CHECK after `/reload-plugins`: best-tier phase w/ Fable out of credits → `/route show` lists fable down (probe), main on opus, effort bumped. -- [ ] LIVE CHECK: does `parentModel` follow the turn.step rewrite or the session model? Read general-purpose's step-0 `message.model` in a transcript during a demoted turn (if it is already the session model, the pin is a harmless no-op). -- [ ] W1-C live verification part 2 (needs a real incident): StopFailure vs turn.complete order; `PostModelSwitch` `auto` semantics and its `from_model` after a router upgrade; whether a router rewrite raises `auto`; `[1m]` carry validity on opus (only on a session whose steps carry it) -- [ ] W4-B AC9: type a few plain prompts, read `route: classifier ` (verbose line with `/route verbose on`), drop `bookkeeping` if it misfires +- [x] Fable out of credits: closed "not reproducible, reopen on incident" (block I 2026-11-14; Fable answered all session, 5.1 then 5.2). +- [x] parentModel: not logged (subagent meta = `model` + `pickedEffortModel` only); 240 spawns this session all pinned, none unpinned in a demoted turn → not reproducible from logs, reopen when one runs (block I 2026-11-14). +- [x] W1-C part 2 (StopFailure order, PostModelSwitch `auto`): closed "not reproducible, reopen on incident" (block I 2026-11-14, no incident in this run). +- [x] W4-B AC9: user step, not Claude-runnable (the `route: classifier` line is UI-only, not persisted): `/route verbose on` + a few plain prompts; a misfiring `bookkeeping` → reopen (block I 2026-11-14). - [ ] W3-A: a parallel same-agent dispatch answered after > 10 s; what an unanswered dialog resolves to on the terminal (must be Later or nothing written). T1 skill-row dialogs seen live (f8665b2, d3635b6) - [ ] cold-read cap on the opus session: a coding prompt → spinner `sonnet-5-5/medium`; above 120k cold read the skip line once; `/route show` cap text