Issue Implement

Workflow

Step 1: Retrieve Context

1. Find issue ID from conversation history (from previous issue-plan or issue-implement invocations)

   • If no issue ID found: Ask user "Which issue would you like to continue working on? Please provide the issue ID."
   • Exit and wait for user response if issue ID not found

2. Locate deep-dive artifacts in /tmp/deep-dive/{task-name}/

   • Find the directory associated with this issue from conversation context
   • If multiple directories exist and association is unclear, ask user to confirm
   • Verify these files exist:
     • research.md - Codebase analysis and technical constraints
     • innovate.md - Chosen approach and reasoning
     • plan.md - Implementation steps to follow

Step 2: Fetch Latest Updates

Use gh issue view {issue-id} --json title,body,comments,labels to get all comments since last interaction.

Step 3: Remove Pending Label

Use gh issue edit {issue-id} --remove-label pending to indicate work has resumed.

Step 4: Analyze Feedback

Review new comments for:

• Plan approval/rejection
• Modification requests
• Additional requirements
• Questions or clarifications

Step 5: Take Action Based on Feedback

• If plan approved: Proceed to implementation (Step 6)
• If changes requested:
  • Update /tmp/deep-dive/{task-name}/plan.md
  • Post revised plan as comment
  • Add "pending" label and exit
• If questions asked:
  • Answer questions in comment
  • Add "pending" label and exit

Step 6: Implementation

1. Read deep-dive artifacts:

   • Read plan.md for the exact implementation steps to follow
   • Reference research.md for codebase understanding and navigation
   • Reference innovate.md for the chosen approach and its rationale

2. Create/switch to feature branch

3. Implement changes following plan.md exactly:

   • Follow the implementation steps in order
   • For new user-facing features, register a FeatureSwitchKey and gate the feature — consult /feature-switch skill for step-by-step instructions
   • Do not deviate from the approved plan without user approval
   • If plan is unclear or needs adjustment, post comment and add "pending" label

4. Write and run tests after each change

5. Commit with conventional commit messages

Step 7: Create PR

• If work complete: Push branch and create Pull Request. Report PR URL.
• If blocked or need clarification: Post comment explaining the situation, add "pending" label, and exit
• If intermediate checkpoint: Post progress update comment, add "pending" label, and exit (optional)

Label Management

• Remove "pending" label when resuming work (Step 3)
• Add "pending" label when:
  • Waiting for plan approval (revised plan)
  • Blocked and need user input
  • Optional: intermediate progress checkpoints

Error Handling

• If issue ID cannot be found in conversation context: ask user to provide issue ID and exit
• If deep-dive artifacts not found: ask user if they want to run issue-plan skill first
• If "pending" label doesn't exist: create it first with gh label create pending --description "Waiting for human input" --color FFA500
• If tests fail during implementation: report failures, add "pending" label, and ask for guidance