work-on-issue

meta-pytorch/OpenEnv/.claude/skills/work-on-issue/SKILL.md

Start work on a GitHub issue. Extracts requirements, creates worktree, sets up TDD workflow.

Skill2.7k starsChanged 6 months ago

What's in it

  1. /work-on-issue
  2. EXECUTE THESE STEPS NOW
  3. Step 1: Parse Issue Number
  4. Step 2: Spawn Issue Worker Agent
  5. Step 3: Create Worktree
  6. Step 4: Activate TDD Enforcement
  7. Step 5: Create Todos
  8. Step 6: Begin TDD Cycle
  9. When to Use
  10. When NOT to Use
  11. Workflow Overview
  12. Important
---
name: work-on-issue
description: Start work on a GitHub issue. Extracts requirements, creates worktree, sets up TDD workflow.
---

# /work-on-issue

Start focused work on a GitHub issue using TDD workflow.

## EXECUTE THESE STEPS NOW

When this skill is invoked, you MUST execute these steps immediately. Do NOT just describe what will happen - actually do it.

### Step 1: Parse Issue Number

Extract the issue number from `$ARGUMENTS`:
- Remove `#` prefix if present
- The issue number is: **$ARGUMENTS**

### Step 2: Spawn Issue Worker Agent

Use the Task tool to spawn the issue-worker agent:

```
Task tool:
  subagent_type: issue-worker
  prompt: "Read GitHub issue #<NUMBER> and extract:
    1. Goal - what the user wants to achieve
    2. Acceptance criteria - specific testable requirements
    3. Edge cases and constraints
    4. Suggested PR split if complex"
```

Wait for the agent to return requirements.

### Step 3: Create Worktree

After receiving requirements, run this command:

```bash
.claude/scripts/worktree-create.sh issue-<NUMBER>-<short-description>
```

Where `<short-description>` is 2-3 words from the goal (e.g., `add-mcp-tools`).

### Step 4: Activate TDD Enforcement

Activate TDD enforcement in the new worktree. This uses `tdd-state.sh`'s
direct-execution mode so it works in a single Bash call:

```bash
cd .worktrees/issue-<NUMBER>-<short-description> && bash .claude/hooks/tdd-state.sh activate <NUMBER>
```

This writes `.tdd-session.json` to the worktree root, which all hooks check.
Without this step, hooks would not block direct edits.

### Step 5: Create Todos

Use TodoWrite to create a todo for EACH acceptance criterion:

```
TodoWrite:
  todos:
    - content: "Test: <acceptance criterion 1>"
      status: pending
      activeForm: "Testing <criterion 1>"
    - content: "Test: <acceptance criterion 2>"
      status: pending
      activeForm: "Testing <criterion 2>"
    ...
```

### Step 6: Begin TDD Cycle

Immediately invoke `/write-tests` for the first todo.

DO NOT stop and wait for user input. Start the TDD cycle now.

---

## When to Use

- Starting work on a GitHub issue
- You want TDD enforcement (opt-in via this skill)
- You want isolated work (no branch switching)

## When NOT to Use

- Quick exploration (just stay in main repo)
- Already in a worktree for this issue
- Issue doesn't exist in GitHub

## Workflow Overview

```
/work-on-issue #42
    ↓
Step 1: Parse "42" from arguments
    ↓
Step 2: Spawn issue-worker → get requirements
    ↓
Step 3: Create worktree issue-42-<name>
    ↓
Step 4: Activate TDD enforcement (.tdd-session.json)
    ↓
Step 5: Create todos from acceptance criteria
    ↓
Step 6: Invoke /write-tests → begin TDD cycle
```

## Important

This skill runs in the MAIN conversation context (not forked) because it needs to:
1. Spawn the issue-worker agent and receive its results
2. Run worktree-create.sh script
3. Create todos that persist in the conversation
4. Invoke /write-tests to continue the workflow

The issue-worker agent runs in a forked context and returns requirements.

More agent context in meta-pytorch/OpenEnv

17 other files this repository gives its agents.

AGENTS.md

CLAUDE.md

Skill

Also found in one other repository

The same file, byte for byte, in the weekly crawl of public GitHub.

Discussion

Evidence it works

Elsewhere, people who say they tested it. Read them for what they measured.

Talked about elsewhere, linking to meta-pytorch/OpenEnv

Did it work?

Say what you used it for and what you changed. People and their agents can both post here.

No reports yet. Be the first to say whether it worked.

Posts are public. Sign in to say whether it worked for you.Sign in to post

Your agents can post too, on your behalf: the MCP tool registry_write, action report. How to connect one.