Worktree Plan

1. Read the full issue and design

gh issue view <issue_number> --json title,body,labels,comments

• Parse the body for acceptance criteria and constraints.
• Find the ## Design comment for the approved architecture.
• If an ## Implementation Plan comment already exists, skip — the planning phase is done.
• If no ## Design comment exists, invoke worktree_brainstorm first.

2. Break into tasks

• Each task should be completable in 2-5 minutes.
• Each task must specify:
  • What: one sentence describing the change.
  • Files: exact file paths to create or modify.
  • Verification: how to confirm the task is done (e.g. just test, specific test passes).
• Order tasks by dependency — earlier tasks must not depend on later ones.
• Follow TDD: test tasks come before or alongside implementation tasks.

3. Publish plan as a GitHub issue comment

1. Determine the repo: gh repo view --json nameWithOwner --jq '.nameWithOwner'

2. Post the plan comment:

   gh api repos/{owner}/{repo}/issues/{issue_number}/comments \
     -f body="<plan_content>"
   

3. The comment must start with ## Implementation Plan (H2) — this is how other skills detect that the planning phase is complete.

4. Use this format:

   ## Implementation Plan
   
   Issue: #<issue_number>
   Branch: <branch_name>
   
   ### Tasks
   
   - [ ] Task 1: description — `files` — verify: `command`
   - [ ] Task 2: description — `files` — verify: `command`
   ...
   

4. Proceed to execution

• Invoke worktree_execute to start implementing.

Delegation

The following steps SHOULD be delegated to reduce token consumption:

• Steps 1, 4 (precondition check, read issue/design): Spawn a Task subagent with model: "fast" that validates the branch name, executes gh issue view, checks for existing ## Design and ## Implementation Plan comments. Returns: issue number, parsed body/design, plan-exists flag.
• Step 3 (publish plan): Spawn a Task subagent with model: "fast" that takes the formatted plan content and posts it via gh api. Returns: comment URL.
• Step 4 (invoke next skill): Can remain in main agent (simple skill invocation).

Step 2 (break into tasks) should remain in the main agent as it requires task decomposition and dependency analysis.

Reference: subagent-delegation rule

Important Notes

• Never block waiting for user input. Make reasonable task breakdowns and move on.
• If a task is too large to describe in one sentence, split it.
• The plan comment is the single source of truth — no local plan files.