pr-description
PR descriptions. Use whenever creating or editing a PR body; UI-affecting changes include a conditional UI Review Handoff.
- License 许可
- MIT
- 所属仓库
- Noahlw/efcc
- 最近更新
- 2026-11-09
- Stars
- ★ 0
- Forks
- 1
01 前言 Front Matter
- name
- pr-description
- description
- PR descriptions. Use whenever creating or editing a PR body; UI-affecting changes include a conditional UI Review Handoff.
03 SKILL.md 正文
PR description template
PR resolve issue(s) must auto-close on merge via GitHub closing keywords.
Rule
First line of PR body must be closing statement, before ## Summary:
Closes #<N>[, #<M>, ...]
- List every issue this PR fully resolves. Use
Closes(notFixes/Resolves— pick one verb, keep consistent across repo). - Only list issues actually resolved by this PR's diff. Issue referenced but deliberately deferred (see ADR marking out of scope) goes in prose, not closing line.
- No issues resolved (docs-only or infra-only PR) → omit line entirely, don't write
Closeswith nothing after it.
UI Review Handoff
Include this section only for a materially app-facing UI or presentation PR. Backend-only, infrastructure-only, and documentation-only PRs omit it.
Candidate
- Review level: L1 / L2 / L3
- Baseline Story: <direct Storybook/Playground locator>
- Material Stories: <only the useful states for this change, or
None> - Risk viewport/platform: <only if relevant, or
None>
Agent self-review
- [ ] Baseline inspected across W7.
- [ ] Material states inspected at relevant risk viewports.
- [ ] Obvious defects found during self-review were repaired.
- [ ] No known in-scope visual defect remains.
- [ ] Relevant focused verification is green.
Human review focus
- <only genuine subjective/high-risk review points; use
Routine owner spot-checkfor ordinary L1 work>
Visual comparison
- <for major visual changes, include useful before/after or representative phone/desktop/risky-state images when available; otherwise omit this subsection>
Live Storybook is the primary visual review surface; images support the review. Do not duplicate the Test plan. L1 means normal AI-owned implementation plus mandatory owner Storybook spot-check; L2 is human design review triggered by design risk; L3 is actual platform/device/assistive-technology/manual truth. Detailed escalation law remains in the central repository authority.
Similarity to master (LLM-judged)
Every PR body MUST include a Similarity to master section. It is a reviewer-facing communication aid that summarizes how much of the existing behavior/output the PR preserves relative to the mainline. It is an LLM judgment, not a measurement — no scripts, no byte-diff, no pixel-diff. The writing agent reads the actual diff and reasons to each verdict.
Rules
- Baseline is explicit. Compare the PR head against
master(resolved via merge-base). Title the section with the reference, e.g.## Similarity to master (vs master @ <sha>). Never report a number without stating what it is compared against. For a stacked PR chain (A → B → C), compare the accumulated head against master — reviewers see what actually lands on the mainline, not just this PR's own delta. If the PR's own delta is materially different from the cumulative tip, say so in one line. - Axes are agent-chosen per diff. Do NOT use a fixed axis list. Read the diff and
pick the axes that are actually informative for this change. Examples of axes you
may choose (add or drop as the diff demands):
- Visual / DOM — UI-rendering source changed? (components, styles, markup)
- Report / output schema — the persisted/emitted contract changed?
- Report values — the data changed? (scores, costs, computed fields)
- Behavior / lifecycle — interactions, status transitions, retry/abort/cancel?
- Config / infra — CI, tooling, dependencies, deployment?
- Docs / ADR — documentation-only, no runtime impact? Drop an axis that the diff doesn't touch rather than reporting a meaningless number. This keeps the report project-agnostic: it works for any repo, not just this one.
- Every verdict is an LLM judgment with a reason. For each axis, give a similarity percentage AND a one-line justification grounded in the real diff (quote the files that moved). The % is an estimate, not a measurement — its authority comes from the reasoning, not the number.
- Honest vocabulary. Use these labels per axis:
unchanged (100%)— the diff does not touch this axis.<N>%— the diff partially changes this axis; justify the estimate.n/a— the axis is nondeterministic output (e.g. LLM-generated prose) and a similarity % would be a fake precision. Say so explicitly.changed by design — see summary— an intentional behavior/data change; link it to the Summary bullet, don't dress it as a similarity failure. Never invent a fake % for nondeterministic content. Never blend axes into one aggregate number.
- Label the report as a judgment. One line at the end: "Similarity is an LLM assessment of the diff, not a CI gate or measurement." This prevents anyone from mistaking it for a hard verification.
Template
## Similarity to master (vs master @ <sha>)
This is an LLM judgment of the diff, not a measurement.
- **<Axis 1>**: <verdict> — <reason, quoting the diff>
- **<Axis 2>**: <verdict> — <reason>
- **<Axis 3>**: <verdict> — <reason>
Similarity is an LLM assessment of the diff, not a CI gate or measurement.
Template (full body)
Closes #<N>, #<M>
## Summary
- <what changed and why, 1-3 bullets>
## Similarity to master (vs master @ <sha>)
- **<Axis 1>**: <verdict> — <reason>
- **<Axis 2>**: <verdict> — <reason>
Similarity is an LLM assessment of the diff, not a CI gate or measurement.
## Test plan
- [ ] <how this was verified>
## QA Test Steps
1. <how qa test this>
🤖 Generated with [Claude Code](https://claude.com/claude-code)
https://claude.ai/code/session_017QWhHHbUxYJ3kusvFbqUDu
Example
Closes #9, #10, #15, #16
## Summary
- primaryColor/welcomeMessage/avatarUrl wired into rendering per ADR-0007
- destroy() added for SPA cleanup
- #17 intentionally left open — deferred per ADR-0008, not part of this PR
## Similarity to master (vs master @ 1a2b3c4)
- **Visual / DOM**: unchanged (100%) — no .tsx/.css files in the diff.
- **Report values**: changed by design — color tokens and welcome copy differ; see Summary.
- **Config**: n/a — dependency/dep-only changes, no runtime surface moved.
Similarity is an LLM assessment of the diff, not a CI gate or measurement.
## Test plan
- [x] bun run typecheck passes
- [x] verified live in headless Chromium
🤖 Generated with [Claude Code](https://claude.com/claude-code)
https://claude.ai/code/session_017QWhHHbUxYJ3kusvFbqUDu
04 同仓库更多 Skills
writing-plans
Use when you have a spec or requirements for a multi-step task, before touching code
writing-for-agents
Writing documents for agents. Use when creating or editing skills, or modifying AGENTS.md or CLAUDE.md.
to-tickets
Break a plan, spec, or the current conversation into a set of tracer-bullet tickets, each declaring its blocking edges, published to the configured tracker — edges as text in one file per ticket locally, or native blocking links on a real tracker.
to-spec
Turn the current conversation into a spec and publish it to the project issue tracker — no interview, just synthesis of what you've already discussed.
tdd
Test-driven development. Use when the user wants to build features or fix bugs test-first, mentions "red-green-refactor", or wants integration tests.
implement
Implement a piece of work based on a spec or set of tickets.