メニュー
Read a GitHub Issue and run branch creation through implementation, commit, and PR. Use when the user asks to implement, work on, or execute a GitHub issue (e.g. "implement #123", "work on Issue #42").
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | implement-issue |
| description | Read a GitHub Issue and run branch creation through implementation, commit, and PR. Use when the user asks to implement, work on, or execute a GitHub issue (e.g. "implement #123", "work on Issue #42"). |
Concrete steps for the flow in ai-workflow.mdc (plan → approval → implement → PR).
gh issue view <number>
Understand background, requirements, and acceptance criteria.
git fetch origin && git checkout -b cursor/<short-description>-<issue-number> origin/main
Do not commit directly to main.
After approval, break down tasks and execute. Follow coding-guide.
git add <files>
git commit -m "<message>"
git log --oneline -10closes #<issue-number> when appropriategit push -u origin HEAD
After push, create the PR per create-pull-request.
timeout and retry. Do not use --no-verify.Periodic code health checkup for design drift not caught by CI. Use at version bumps or when asked for a code health check / 健康診断 / 健診. Audits logged advisory deviations and runs a deepen-modules pass.
Perform a project release: bump the semver `version` in package.json, summarize changes since the last release, and hand off to the health-checkup. Use when cutting a release, bumping the project version, or asked to "version up".
Project coding conventions and architecture constraints. Use when implementing, creating, or modifying components and modules (e.g. "create a component", "implement this feature").
Create a PR aligned with the AI-led workflow: summary, design (when relevant), impact and risks. Use when creating a pull request, opening a PR, pushing for review, or completing implementation of an issue.
Triage GitHub PR review feedback (top-level, inline, threads); implement agreed items; escalate disagreements or questions. Includes lint/test checks, per-thread replies with commit SHA, and a chat summary. Use when fixing review feedback on an open PR—not for greenfield implementation only.
Propose codifying reusable lessons from trial-and-error or explicit user requests into lint, rules, or skills. Use near task completion after failed first attempts, or when the user asks to "make it a rule" or "leave a lesson". NOT for empirically testing instructions on fresh agents (tune-prompt) or drafting a new skill without a lesson to codify.