[wukongnotnull]

Closes #49

This PR MUST target the main branch. Active work is developed on
focused branches and merged directly into main after review.

Who is submitting this PR? (required)

Field	Value
Your model + version	Cursor Grok 4.6
Harness + version	Cursor cloud agent (this run)
All plugins installed	none beyond the Cursor Cloud Agent environment
Human partner who reviewed this diff	悟空非空也g — said 合入 (explicit approval to open and merge this branch)

What problem are you trying to solve?

Plugin users of Wukong Code read skills/grilling/SKILL.md, not this repository’s AGENTS.md. After grilling, agents were dumping the in-chat Shared-Understanding Record into docs/wukong-code/specs/ as if that eight-section confirmation record were the spec (the #32 contract), while this repo’s AGENTS.md later forbade writing specs at all (#50) and put tracking on GitHub issues. 悟空非空也g, in a live session, rejected both: the confirmation record is not the spec; after the record is confirmed, write a separate technical specification derived from it, have the human review that spec file, then writing-plans. Issue/board tracking stays a habit of this repo in AGENTS.md / CLAUDE.md, not a skill capability.

This is not a theoretical cleanup. The failure mode was agents treating the interview transcript dump as the design document, or skipping a spec because AGENTS.md said “issue not spec.”

What does this PR change?

skills/grilling/SKILL.md HANDOFF now writes a derived six-section technical spec (not a paste of the eight Completion Gate headings), waits for spec-file review, and loads writing-plans only after that file is approved. AGENTS.md / CLAUDE.md grilling section is slimmed to this-repo issue tracking and a pointer at the skill. Resident probes in tests/skills/grilling-scenarios.md and a Cursor isolated RED/GREEN note under docs/wukong-code/evals/2026-10-13-grilling-derived-spec-handoff.md match the new contract.

Is this change appropriate for the core library?

Yes. grilling is already a core process skill. Plugin users in any project that loads the skill need the HANDOFF sequence; they will never read this repo’s AGENTS.md. Scope, provenance, and maintenance stay with Wukong Code process skills. Zero new dependencies, no third-party runtime, no GitHub Issues/Projects in the skill. Evidence is the isolated RED/GREEN below, proportionate to a HANDOFF wording change (not a new skill).

What alternatives did you consider?

• Put the derived-spec sequence only in AGENTS.md and leave the skill dumping the record (feat: persist confirmed grilling records as specs before writing-plans #32). Rejected by 悟空非空也g: future plugin users read the skill.
• Never write a spec; issue is the system of record (docs: add GitHub issue and Projects collaboration rules #50 / earlier AGENTS.md). Rejected: confirmation record and issue tracking are not the spec; a separate technical specification is required before writing-plans.
• Put GitHub issue/Projects tracking into skills/grilling/SKILL.md. Rejected: that is this repository’s development habit only.
• Rewrite the record into brainstorming-style alternatives/narrative. Rejected: derive a technical spec from confirmed content only; do not reopen closed branches.

Does this PR contain multiple unrelated changes?

No. Skill HANDOFF, matching scenario text, this-repo AGENTS slim so it does not contradict the skill, and the eval note for that HANDOFF change. No harness, language pack, or unrelated docs.

Existing PRs

• I have reviewed all open AND closed PRs for duplicates or prior art
• Related PRs: feat: persist confirmed grilling records as specs before writing-plans #32, docs: add GitHub issue and Projects collaboration rules #50, feat: add programming grilling skill #13

#32 (merged) persisted the confirmed eight-section record as the spec. This PR keeps persist-then-file-review-then-writing-plans but forbids treating the record as the spec and requires a derived technical specification. #50 (merged) recorded grilling consensus on issues and forbade new specs in this repo; that AGENTS override is removed so it does not fight the skill. #13 added grilling. No open PR covers this derived-spec HANDOFF.

Environment tested

Harness (e.g. Claude Code, Cursor)	Harness version	Model	Model version/ID
Cursor Cloud Agent isolated general-purpose subagents	this cloud-agent run	Cursor Grok 4.6	Cursor Grok 4.6
wukong-code-evals Drill / Claude Code tmux	not run	—	no grilling scenarios in that harness; no Drill API key in this environment

New harness support (required if this PR adds a new harness)

This PR does not add a new harness. Transcript below is N/A.

Clean-session transcript for "Let's make a react todo list"

N/A — this PR does not add harness support. No clean-session bootstrap transcript.

Evaluation

• What was the initial prompt you (or your human partner) used to start the session that led to this change?
  悟空非空也g decided grilling HANDOFF belongs in the skill: 确认记录 → 根据确认后的记录写 spec 技术规格文档（不要把确认记录自己当成 spec 文档）→ 审核 spec 文档 → 再 writing-plans.
• How many eval sessions did you run AFTER making the change?
  Three isolated GREEN probes (M4/M5 derived spec + file review; M6 按这个写计划 then writing-plans; M7 early-stop). Matching RED: M4 dumped the eight-section record as the spec; M7 already passed (no file). Official wukong-code-evals grilling cohort: not run (repo has JS/TS jsonl only on the language-guidance branch; no API key). Do not treat this as a Drill/tmux result.
• How did outcomes change compared to before the change?
  Before: HANDOFF copied Objective…Success Criteria into the spec file. After: spec used Problem and objective / Architecture / Interfaces / Constraints / Errors / Acceptance; asked for review of that file; writing-plans only after spec-file approval; early-stop still wrote nothing. Details: docs/wukong-code/evals/2026-10-13-grilling-derived-spec-handoff.md.

Language-pack evidence

• Language and support status: N/A — not a language pack
• Real failure generic methodology handled incorrectly: N/A
• Project markers and toolchain versions: N/A
• Official sources: N/A
• No-guidance repetitions and failures: N/A
• With-guidance repetitions and results: N/A
• Monorepo, unsupported-language, and docs-only controls: N/A
• Human reviewer familiar with the language: N/A
• Frameworks or third-party preferences introduced: none

Rigor

• If this is a skills change: I used wukong-code:writing-skills and
  completed adversarial pressure testing (paste results below)
• This change was tested adversarially, not just on the happy path
• I did not modify carefully-tuned content (Red Flags table,
  rationalizations, "human partner" language) without extensive evals
  showing the change is an improvement

RED (current skill, new contract): M4 actor bc-6dd09e92-fff4-5d84-aed3-c4757f76884a wrote a spec whose outline was the eight Completion Gate headings. M7 actor bc-30288003-776b-5222-acb1-cd0b217b4290 stopped with a partial record and wrote no file.

GREEN (candidate skill): M4/M6 actor bc-89eb817f-89b3-53d1-a6d1-d558430a991e wrote /tmp/grilling-green-iso-handoff/docs/wukong-code/specs/2026-10-13-1205-in-process-request-deduplication-design.md with the six derived sections, reported commit failure (isolated tree is not a git repo), asked for spec-file review; on 按这个写计划 loaded writing-plans. M7 actor bc-475400e1-f5b9-546a-850c-e4f3d0d75a24 wrote none.

S1–S5 Turn Contract and U1 router probes were not re-run (HANDOFF-only). Isolated /tmp trees, not session-start bootstrap.

Human review

• A human has reviewed the COMPLETE proposed diff before submission

悟空非空也g reviewed the work on branch cursor/grilling-spec-handoff-9b62 and said 合入.