[wukongnotnull]

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 (cursor-grok-4.6-xhigh-fast)
Harness + version	Cursor Cloud Agent (implementation run bc-7dcec1b4-f2a4-501e-a6d5-52a95659918f; Project coordinator bc-00bc4604-2af3-4dc7-9e8c-350a7f19ede4)
All plugins installed	GitHub MCP, cursor-cloud, cursor-subscriptions, Figma, Lark, X, frontend-design, x-api-mcp-guide, env-setup
Human partner who reviewed this diff	悟空非空也. Reviewed the complete commit (7cc09ade441962074ac64bed018ef0fc37734946, 16 files, +150/−15) in the Project chat on 2026-10-04 and then said to open this PR.

What problem are you trying to solve?

14 process skills had no agents/openai.yaml. scripts/package-codex-plugin.sh filled those files from a prior official pack (--metadata-source) and died if a skill was missing there. CI stayed green by writing those 14 yaml files in tests/codex/test-package-codex-plugin.sh (write_metadata_fixture()). This checkout alone could not cut a complete Codex zip.

This is the leftover architecture item 3, reconfirmed on origin/main bcadfde (v6.4.0) in the 2026-10-04 optimization review. Not a live Codex portal upload failure. The user experience: a new process skill, or a clean checkout with no _tmp/sup-codex-packaging/, cannot ship.

The 14 skills: brainstorming, dispatching-parallel-agents, executing-plans, finishing-a-development-branch, receiving-code-review, requesting-code-review, subagent-driven-development, systematic-debugging, test-driven-development, using-git-worktrees, using-wukong-code, verification-before-completion, writing-plans, writing-skills.

What does this PR change?

Adds one minimal agents/openai.yaml per missing process skill (interface.display_name / short_description / default_prompt + policy.allow_implicit_invocation: true), matching the existing 12 files. package-codex-plugin.sh no longer requires --metadata-source when the source tree is complete. Sync still preserves destination-side yaml. No SKILL.md edits.

Is this change appropriate for the core library?

Yes. This is core Codex publish metadata, not a new skill.

• Who has the problem: anyone packaging this plugin for Codex from this checkout.
• Why core instead of a plugin: the 14 skills and package-codex-plugin.sh already live in core. Missing yaml is why the packer needs an external tree.
• Externally sourced: no new skill. Field shapes follow the 12 yaml files already in this repo (grilling, language-guidance, 10 product-design*). Official Superpowers yaml in openai/plugins is still two fields (display_name + short_description); this checkout’s existing files already add default_prompt and implicit policy.
• Evidence: tests/codex/test-package-codex-plugin.sh (source yaml wins over fixture; no-metadata-source archive matches the fixture run; incomplete fallback still rejected if writing-plans yaml is omitted) and tests/codex-plugin-sync/test-sync-to-codex-plugin.sh passed. After commit, package-codex-plugin.sh --output … with no --metadata-source produced 26 skills / 26 yaml. No Codex portal upload.

What alternatives did you consider?

1. Keep requiring --metadata-source and only document it. Rejected: that is the current failure mode; CI hides it with fixtures.
2. Generate yaml at pack time from SKILL.md and do not commit the files. Rejected: OpenAI metadata is their contract; the existing 12 files are committed, and pack/sync already treat committed yaml as source of truth.
3. Copy the official Superpowers two-field yaml only. Rejected: this repo’s 12 files already use four fields; mixing shapes would split the contract.
4. Also fold hook archive paths into codex-package.manifest.json. Out of scope (optimization item 3 in the 2026-10-04 list).

Does this PR contain multiple unrelated changes?

No. The 14 yaml files, the packer change that makes a complete tree self-sufficient, and the package tests that lock “no --metadata-source when source is complete” are one metadata-completeness fix. No skill prose, PD bootstrap, or short-description work.

Existing PRs

• I have reviewed all open AND closed PRs for duplicates or prior art
• Related PRs: #36, #13, #11, #12

No open PR on cursor/process-openai-yaml-918f. No closed PR commits these 14 files. #36 only unified PD transport paths. #13 added grilling yaml only. #11 / #12 added language-guidance and the “source yaml wins over fixture” packer behavior.

Environment tested

Harness (e.g. Claude Code, Cursor)	Harness version	Model	Model version/ID
Cursor Cloud Agent (static: tests/codex/test-package-codex-plugin.sh, tests/codex-plugin-sync/test-sync-to-codex-plugin.sh; post-commit package-codex-plugin.sh zip with no --metadata-source)	Cursor Cloud Agent, 2026-10-04	Cursor Grok 4.6	cursor-grok-4.6-xhigh-fast
Live Codex portal upload	not run	—	—

scripts/lint-shell.sh was not run (shellcheck not on PATH). bash -n on touched shell.

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

Not applicable. This PR does not add a new harness.

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

Not applicable — no new harness integration.

Evaluation

• What was the initial prompt you (or your human partner) used to start
  the session that led to this change?
  After a 2026-10-04 optimization review of origin/main, the human partner accepted “先做第 1 条”: commit minimal agents/openai.yaml for the 14 process skills so this checkout can package Codex without an external pack.
• How many eval sessions did you run AFTER making the change?
  Zero live harness or portal sessions. After the change we ran the Codex package and sync suites above, then packed a zip from the committed tree.
• How did outcomes change compared to before the change?
  Before: 14 skills missing yaml; packer required a prior official pack; CI invented those files. After: 26/26 yaml in tree; packer succeeds with no --metadata-source; tests lock source-complete and incomplete-fallback paths. Sync still keeps destination-owned yaml. We did not upload to the Codex portal. using-wukong-code yaml uses this repo’s name; using-git-worktrees follows this fork’s “native isolation or worktree” wording.

Language-pack evidence

• Language and support status: not a language pack
• Real failure generic methodology handled incorrectly: not applicable
• Project markers and toolchain versions: not applicable
• Official sources: not applicable
• No-guidance repetitions and failures: not applicable
• With-guidance repetitions and results: not applicable
• Monorepo, unsupported-language, and docs-only controls: not applicable
• Human reviewer familiar with the language: not applicable
• 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

This is Codex skill metadata and packer fallback, not a skill-prose rewrite. writing-skills was not used. No SKILL.md / Red Flags / “human partner” edits. Coverage is the package/sync suites plus a real no-metadata-source zip. We did not run live multi-session pressure tests.

Human review

• A human has reviewed the COMPLETE proposed diff before submission