Subagent-Driven Development

When This Skill Activates

• •An approved implementation plan exists
• •User says "go", "execute", "start building"
• •The /execute command is invoked

Prerequisites

• •An approved plan MUST exist. If not, activate writing-plans first.
• •The plan must be saved in .factory/docs/plans/.

Process

For Each Task in the Plan:

Step 1: Delegate to Task Executor

Task tool → subagent_type: task-executor
Provide: The single task details, project context

Step 2: Check Executor Result

• •If ✅ PASS → proceed to review
• •If ❌ FAIL → delegate to debugger droid:
  code

  Task tool → subagent_type: debugger
  Provide: The failure details, file paths, error messages
  

  After debug fix, re-run the task executor.

Step 3: Two-Stage Review

Task tool → subagent_type: code-reviewer
Provide: The spec, changed files, repo path

• •If Stage 1 (correctness) BLOCKS → fix and re-review
• •If Stage 1 PASSES → note Stage 2 (quality) suggestions for later

Step 4: Report Progress

After each task, report to the user:

• •Task N of M completed ✅
• •Brief summary of what was done
• •Any reviewer suggestions noted

Step 5: Commit

After a verified, reviewed task:

• •Stage the changed files
• •Commit with a descriptive message: feat(component): task N — description

After All Tasks Complete:

Final Verification

Task tool → subagent_type: verifier
Provide: Original spec, plan, all changed files, repo path

Report the final verification result to the user.

Error Recovery

• •If a task fails after 2 attempts → pause, report to user, ask for guidance
• •If the debugger can't find the root cause → escalate to user with full evidence
• •Never silently skip a failing task

Output

A fully implemented, tested, reviewed, and verified feature.