fix(hooks): merge end-of-turn hook — recap via reason (supersede #81), 3 branches + visible marker - #89
Merged
Merged
Conversation
…ede #81), 3 branches + visible marker #81 (cfa28a3, on main) delivered the recap INSTRUCTION via systemMessage, which is user-only and never reaches the model — verified dual-channel (CC hooks docs Stop decision-control: "reason must be provided for Claude to know how to proceed" + WebSearch). So #81's recap prompt was never seen by the model. d695ac5/redesign correctly use `reason` (injected; agent gets one more turn). This merges best-of-all-three: - reason-delivery (model receives) + short systemMessage human note [supersede #81 delivery bug] - forced visible «## 🟢 Простыми словами» header [from #81 — human scannability] - session-goal anchor (aiTitle→first-msg fallback) + 3-way drift verdict + recommendation-first [from redesign] - 3 branches incl. combined long+question (keeps recommendation-first when a fork is on the table) - stateless-fire preserved (no /tmp, no aggregation/isMeta) Tested 2026-05-21: bash -n; branches A/B/C; silent on bare tool_use; AskUserQuestion→B; anchor aiTitle-primary + first-user-msg fallback + graceful default; §1.8 skip smoke-test exit 0. Prior-art: skipped — modifies existing <50-LOC-logic CC hook (delivery-channel merge), no new capability per dual-implementation-discipline §2(iv); SSOT #8/#9/#20 cover the anchoring/hook family.
artyhoo
added a commit
that referenced
this pull request
May 21, 2026
…#91) Provenance for the hook redesign whose IMPL already shipped (#89 — session-anchored dual-audience recap with aiTitle goal anchor). The design research-patch (D1-D6 locked, Rounds 1-4 + decisions) was committed on feat/eot-hook-anchor + chore/ssot-karpathy-skills-ref but never reached main. Brought as the single file only (verified identical across both branches; no branch legacy) onto a fresh branch off origin/main. Prior-art: skipped — landing an already-authored, approved design research-patch; impl already on main via #89, no new capability introduced.
This was referenced May 21, 2026
Merged
artyhoo
added a commit
that referenced
this pull request
May 21, 2026
Lands the maintainer's settings.json safety-hardening that was stranded uncommitted on the superseded chore/ssot-karpathy-skills-ref branch (the only non-superseded residue there — every other file already on main via #89/#92/#93/ #94/#96/#97/#98). Committed verbatim. On-thesis: this is edit-time (earliest-channel) enforcement of project safety rules that previously lived only as memory/prose — "no git reset hard ever", no force-push, no skip-verify. The harness now blocks them before they run: - destructive git: force/mirror push, push to main/master, skip-verify commit, hard reset, checkout discard, branch force-delete, filter-branch, update-ref - destructive fs: recursive delete, shred, dd, mkfs, sudo - pipe-to-shell: curl-to-shell, wget-to-shell; raw netcat/socat - secret exfil: gh secret set/delete, gh auth logout, reads of credential dirs (ssh / aws / gnupg / gh config) and dotenv files - self-protection: .git/hooks, .husky, settings.json (Edit/Write deny) Also: portable $CLAUDE_PROJECT_DIR hook paths + ask-question-reminder.sh permission. Prior-art: skipped — harness config hardening (permission deny-list), no new capability code or dependency
5 tasks
artyhoo
added a commit
that referenced
this pull request
May 21, 2026
…#106) Append + in-place refinement of the living history book (v1 technical +86 lines, v2 literary +78). Narrates post-#93 events: the end-of-turn hook saga (#81 broken via systemMessage → #89 reason-channel fix), #94 AskUserQuestion hook, principles 11/12/13 landing, the #96/#97 claim-scan experiment (H0 not rejected), #90 recommendation-gate, and the niche-among-companions chapter. Colophon extended (chapters III–XII, patches to 2026-05-21, PR #85–#99). Preserves continuation work that was stranded uncommitted on the superseded chore/ssot-karpathy branch. Prior-art: skipped — narrative doc continuation, no new capability or dependency.
5 tasks
artyhoo
added a commit
that referenced
this pull request
May 31, 2026
…brella done.md (#307) Closes the companion-reuse-deep-dive umbrella tail (PRs #248-251 already merged): - SSOT: 3 new append-only rows #88 (aif-handoff autoQueueMode LOGIC, ADOPT VOCABULARY + REFERENCE), #89 (omo session-plan-affinity, ADOPT VOCABULARY), #90 (Superpowers brainstorming scope-assessment, ADOPT VOCABULARY) — content copied from the merged R-phase patches; bare-ID column per principle-08 loader. #88 verdict-column normalized to clean `ADOPT VOCABULARY` with the faceted `+ REFERENCE` detail moved into the rationale column (principle-11 F2 convention, matches faceted rows #86/#87). - SSOT #67: additive note appended to «Trigger to revisit» only; REJECT verdict for the full Kanban runtime unchanged. - Resolved [next-available-slot] placeholders in the 3 source patches → #88/#89/#90. Devin (Sub-wave C) creates no row — its 2 descriptive placeholder mentions left as-is. - done.md written per CLAUDE.md umbrella-closure schema (Final PR #251). Verified: principle suite 153 pass / 4 skip; markdownlint clean. Prior-art: skipped — doc-only edit (SSOT register rows + umbrella done.md + research-patch placeholder resolution); no new capability, dependency, or code.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What & why
The end-of-turn hook diverged into three versions; #81 (
cfa28a3, currently onmain) andd695ac5made opposite assumptions about which Stop-hook field reaches the model. This PR settles it by evidence and merges the best of all three.Load-bearing finding (verified dual-channel, [[no-paid-llm-in-ci]]-compatible — docs + WebSearch, no API):
decision:"block",reasonreaches the model — CC hooks docs (Stop decision-control): «reason must be provided for Claude to know how to proceed»; «the agent gets one more turn with thereasoninjected». Corroborated by independent WebSearch.systemMessageis user-UI only — does NOT reach the model.Therefore #81 is functionally broken: it delivers the entire recap instruction via
systemMessage(user-only), so the model never receives the prompt to write the recap block. Itsreasonis a static placeholder.d695ac5/redesign correctly usereason(andd695ac5's in-code "systemMessage НЕ доходит" comment is verified-accurate — it was the 1.5h-later fix for #81's delivery bug). This PR supersedes #81's delivery bug.Best-of-both merged (maintainer-confirmed forks 2026-05-21)
reason(model) + shortsystemMessagehuman notecf650d4, D1–D3)Branch C (long answer AND trailing fork-question) carries both the work recap (+ drift verdict) AND recommendation-first/fork-challenge — the 2-branch "long-wins" version dropped the recommendation nudge exactly when a fork is on the table.
Tests (2026-05-21, synthetic JSONL fixtures)
bash -nclean · Branch A (long→work recap, no question section) · Branch B (question→fork-challenge + recommendation-first) · Branch C (long+question→both) · silent on bare tool_use (exit 0) · AskUserQuestion→Branch B · anchor aiTitle-primary + first-user-msg fallback + graceful default · §1.8 skip smoke-test exit 0, no stderr.Not a capability commit
Modifies existing <50-LOC-logic CC hook; no new dep; not under
packages/.@cc-only-rationaleretained. No new SSOT entry (existing #8/#9/#20 cover the anchoring/hook family). Escape-hatchPrior-art:trailer present.§1.7 Self-reflection
Forward-check (what future work must honor)
reason, notsystemMessage(the delivery comment at those lines is load-bearing —systemMessagedoes not reach the model)./tmp) must not regress.Backward-check (existing rules/claims this PR leans on, verified)
SyncHookJSONOutput.reasoncarries no JSDoc routing, so §1.10 step 4 (prose dual-channel) applies; recorded at docs/meta-factory/research-patches/2026-05-21-eot-hook-merge-delivery-semantics.md:30-32.@cc-only-rationaleretained at .claude/hooks/end-of-turn-reminder.sh:2; no portable channel added → no@dual-pairneeded.Post-merge note
chore/ssot-karpathy-skills-ref(cf650d4) still carries the older 2-branch redesign hook. When that branch merges, its hook change will conflict with this one — resolve toward this (newer, 3-branch, reason-delivery) version; do not silently revert.