AgentSkillsCN

work

在就绪的GitHub问题上开展工作（实施、测试、提交、PR）

SKILL.md
--- frontmatter
name: work
description: Work on a ready GitHub issue (implement, test, commit, PR)
argument-hint: [issue-number] (optional)
allowed-tools: Bash, Task, Read, Grep, Glob, Edit, Write
disable-model-invocation: true
agent: work-agent

Work on Issue Command

Implement a ready-labeled GitHub issue: research, implement, review, commit, and open a PR.

Dynamic Context

Current git status: !git status --short 2>/dev/null || echo "Not in a git repo"

Current branch: !git branch --show-current 2>/dev/null || git rev-parse --short HEAD 2>/dev/null || echo "Unknown"

Recent commits: !git log --oneline -20 2>/dev/null || echo "No git history found"

Prerequisites

Issues must have the ready label (groomed with clear description, acceptance criteria, out-of-scope section).

Arguments

Issue number: $ARGUMENTS

Instructions

Step 1: Select Issue

If an issue number was provided above, verify it has the ready label. If not, inform the user and suggest running /groom first.

If no issue number was provided:

  1. •Fetch issues: gh issue list --label ready --json number,title,labels,milestone --limit 20 (exclude any with in-progress label)
  2. •If none found, inform the user and exit
  3. •Present the issues and ask the user to select one

After selection, mark it in-progress: gh issue edit <number> --add-label in-progress

Step 2: Understand the Issue

  1. •Fetch full details: gh issue view <number> --json number,title,body,labels,milestone
  2. •Parse: Description, Acceptance Criteria, Out of Scope
  3. •Present a brief summary showing you understand the requirements

Step 3: Research Phase

  1. •

    Read conventions: Use Read tool on docs/ai/conventions.md

  2. •

    Explore codebase using the explore-issue agent:

    • •Spawn a Task with subagent_type: explore-issue
    • •Find related files, existing patterns, similar implementations
    • •Identify files to create or modify
  3. •

    Identify approach: files to change, patterns to follow, technical decisions, change order

Step 4: Setup Branch

  1. •Fetch latest: git fetch origin main
  2. •Create feature branch from origin/main:
    • •Convention: <type>/<issue-number>-<short-description> (types: feat/, fix/, refactor/, docs/, chore/)
    • •Example: git checkout -b feat/42-user-authentication origin/main
  3. •If branch exists, see templates/edge-cases.md for resolution

Step 5: Implementation

Make changes following the approach from research:

  • •Edit existing files with Edit tool, create new files with Write tool
  • •Keep changes focused on acceptance criteria — avoid scope creep

Core loop — repeat until green:

bash
pnpm check    # lint + build + test (REQUIRED before committing)

Fix any failures before proceeding. For UI changes, also run:

bash
pnpm -F web test:e2e:smoke    # REQUIRED for UI changes

For specialized scenarios (DB migrations, CSS tokens, visual-compare), see templates/implementation-guide.md.

If implementation reveals out-of-scope work, note it for follow-up issues — stay focused on acceptance criteria.

Step 6: Verify Acceptance Criteria

Go through each acceptance criterion from the issue:

  1. •Confirm each item is satisfied by the implementation
  2. •Run relevant tests to verify
  3. •If any criterion is not met, ask the user whether to address now or note for follow-up

Step 7: Self-Review (REQUIRED)

You MUST spawn 3 parallel review agents before committing. DO NOT skip this step or proceed directly to commit.

  1. •

    Get the diff of all changes:

    bash
    git diff --name-only
git diff
  2. •

    Spawn 3 review agents in parallel using Task tool:

    • •subagent_type: conventions-review — check against project conventions
    • •subagent_type: code-quality-review — check for over-engineering and unnecessary complexity
    • •subagent_type: reuse-review — check for code duplication and reuse opportunities

    Pass each agent the changed files list and diff content.

  3. •

    Collect findings from all 3 agents.

  4. •

    If issues found, present them grouped by category and ask the user how to proceed:

    • •Fix all: Auto-fix all identified issues
    • •Fix selected: Let user pick which to fix
    • •Skip: Proceed without fixes (add note to PR)
  5. •

    Apply any approved fixes, then re-run pnpm check to verify.

  6. •

    If no issues found, proceed to Step 8.

Step 8: Commit Changes

GATE CHECK — Before proceeding, confirm Step 7 was completed:

  • •Did you spawn all 3 review agents (conventions, code-quality, reuse)?
  • •Did you collect and process their findings?
  • •Did you apply approved fixes or confirm no issues were found?

If ANY answer is NO, STOP and return to Step 7 now.

  1. •Review changes: git status and git diff
  2. •Stage relevant files (avoid staging unrelated changes)
  3. •Commit with issue reference:
    bash
    git commit -m "<type>: <description> (#<issue-number>)"
  4. •Push branch: git push -u origin <branch-name>

Step 9: Create Pull Request

Create PR using the template from templates/pr-body.md:

bash
gh pr create --title "<title>" --body "$(cat <<'EOF'
<fill in from templates/pr-body.md>
EOF
)"

Include Closes #<issue-number> in the body. Then remove the in-progress label:

bash
gh issue edit <number> --remove-label in-progress

Step 10: Summary

Provide a summary including:

  • •Issue worked on (number and title)
  • •Branch name and PR link
  • •Key changes made and files modified
  • •Any follow-up items identified

Notes

  • •ALWAYS read the issue thoroughly before starting
  • •ALWAYS research the codebase to understand existing patterns
  • •NEVER skip self-review — always spawn review agents before committing
  • •Follow conventions in docs/ai/conventions.md
  • •Keep commits focused and atomic; reference the issue number in commits and PR
  • •For edge cases (issue not ready, lint errors, branch conflicts, etc.), see templates/edge-cases.md