/implement

Orchestrates the full implementation of a planned feature by spawning agents sequentially: backend → frontend → QA → code review.

Prerequisites

A plan file must exist at plans/<feature>/plan.md. This is created by Plan Mode (see .claude/rules/planning.md).

Usage

/implement
/implement <feature-name>

If no feature name is given, find the most recently modified plans/*/plan.md file.

Orchestration Flow

1. Find the Plan

# If feature name provided:
plans/<feature-name>/plan.md

# If no feature name, find most recent:
ls -t plans/*/plan.md | head -1

Read the plan file. Confirm with the user which feature to implement if ambiguous.

2. Spawn Backend Developer

Use the Task tool with subagent_type: laravel-developer:

Prompt: "Read the plan at plans/<feature>/plan.md and implement all backend changes."

Wait for completion. If the agent reports test failures it couldn't fix, stop and report to the user.

3. Spawn Frontend Developer

Use the Task tool with subagent_type: inertia-developer:

Prompt: "Read the plan at plans/<feature>/plan.md and implement all frontend changes. The backend is already implemented — read the code to understand routes and props."

Wait for completion. If the agent reports test failures it couldn't fix, stop and report to the user.

4. Spawn QA Engineer

Use the Task tool with subagent_type: qa-engineer:

Prompt: "Read the plan at plans/<feature>/plan.md. The backend and frontend are already implemented. Run the full test suites, verify acceptance criteria coverage, and audit for security issues."

If QA reports failures:

• Identify whether the failure is backend or frontend
• Re-spawn the appropriate developer agent (laravel-developer for backend, inertia-developer for frontend) with the failure context: describe the specific error, file:line, and what needs fixing
• After the developer agent fixes the issue, re-spawn QA
• Maximum 2 retry cycles. If still failing after 2 retries, stop and report to the user.

5. Spawn Senior Code Reviewer

Use the Task tool with subagent_type: senior-code-reviewer:

Prompt: "Review all code changes for this feature. Run git diff to see what changed. Check code quality, architecture alignment, and security. Provide a verdict: APPROVE, APPROVE WITH SUGGESTIONS, or REQUEST CHANGES."
Model: sonnet
Max turns: 15

6. Report to User

After all agents complete, summarize:

• What was implemented (brief)
• QA result (pass/fail)
• Review verdict
• Any issues that need manual attention

Error Handling

• If any agent hits its max turn limit, report this to the user as a potential sign of a loop
• If backend fails, do not proceed to frontend
• If frontend fails, do not proceed to QA
• If QA fails after 2 retries, skip review and report failures to user
• Always tell the user what happened and what to do next