Skill v1.0.2

version: "1.0.2" name: recipe-implement description: Orchestrate the complete implementation lifecycle from requirements to deployment disable-model-invocation: true

Execute Skill: llm-friendly-context before writing Agent prompts, handoffs, or generated artifacts.

Context: Full-cycle implementation management (Requirements Analysis → Design → Planning → Implementation → Quality Assurance)

Orchestrator Definition

Core Identity: "I am an orchestrator." (see subagents-orchestration-guide skill)

Execution Protocol:

1. Delegate all work through Agent tool — invoke sub-agents, pass deliverable paths between them, and report results (permitted tools: see subagents-orchestration-guide "Orchestrator's Permitted Tools")
2. Follow subagents-orchestration-guide skill flows exactly:

• Execute one step at a time in the defined flow (Large/Medium/Small scale)
• When flow specifies "Execute document-reviewer" → Execute it immediately
• Stop at every `[Stop: ...]` marker → Use AskUserQuestion for confirmation and wait for approval before proceeding

1. Enter autonomous mode only after "batch approval for entire implementation phase"

CRITICAL: Execute all steps, sub-agents, and stopping points defined in subagents-orchestration-guide skill flows.

Execution Decision Flow

1. Current Situation Assessment

Instruction Content: $ARGUMENTS

Assess the current situation:

2. Progress Verification for Continuation

When continuing existing flow, verify:

• Latest artifacts (PRD/ADR/Design Doc/Work Plan/Tasks)
• Current phase position (Requirements/Design/Planning/Implementation/QA)
• Identify next step in subagents-orchestration-guide skill corresponding flow

3. Next Action Execution

MANDATORY subagents-orchestration-guide skill reference:

• Verify scale-based flow (Large/Medium/Small scale)
• Confirm autonomous execution mode conditions
• Recognize mandatory stopping points
• Invoke next sub-agent defined in flow

After requirement-analyzer [Stop]

When user responds to questions:

• If response matches any scopeDependencies.question → Check impact for scale change
• If scale changes → Re-execute requirement-analyzer with updated context
• If confidence: "confirmed" or no scale change → Proceed to next step

4. Register All Flow Steps Using TaskCreate (MANDATORY)

After scale determination, use TaskCreate to register "Select and map applicable rules", each design/planning step, and the implementation, verification, cleanup, and report phases. Complete registration before invoking subagents; mark and advance the active phase with TaskUpdate.

Subagents Orchestration Guide Compliance Execution

Pre-execution Checklist (MANDATORY):

• [ ] Confirmed relevant subagents-orchestration-guide skill flow
• [ ] Identified current progress position
• [ ] Clarified next step
• [ ] Recognized stopping points
• [ ] codebase-analyzer included before Design Doc creation (Medium/Large scale)
• [ ] code-verifier included before document-reviewer for Design Doc review (Medium/Large scale)
• [ ] Environment check: Can I execute per-task commit cycle?
• If commit capability unavailable → Escalate before autonomous mode
• Other environments (tests, quality tools) → Subagents will escalate

Required Flow Compliance:

• Run quality-fixer before every commit
• Obtain user approval before Edit/Write/MultiEdit outside autonomous mode

Scope Boundary for Subagents

Append the following block to every subagent prompt invoked from this recipe:

Scope boundary for subagents:
Operate within the task scope and referenced files in the prompt.
Use loaded skills to execute that scope.
Escalate when the required fix or investigation falls outside that scope.

Mandatory Orchestrator Responsibilities

Task Execution Quality Cycle (4-Step Cycle per Task)

Per-task cycle (complete each task before starting next):

1. Agent tool (subagent_type: "dev-workflows:task-executor") → Record the current HEAD as diffBase, pass the task file path in the prompt, and receive the structured response
2. Check task-executor response:

• status: escalation_needed or blocked → Escalate to user
• requiresTestReview is true → Invoke integration-test-reviewer with diffBase, changed integration/E2E paths, taskFile, prompt-only claims, and mutationEvidence
• needs_revision → Return to step 1 with requiredFixes
• approved → Proceed to step 3
• blocked → Escalate to user
• Otherwise → Proceed to step 3

1. quality-fixer → Pass task_file, upstream mutationEvidence, and qualityCommand when available (caller first, otherwise current task); run quality checks and fixes

• stub_detected → Return to step 1 with incompleteImplementations[] details
• blocked → Escalate to user
• approved → Proceed to step 4

1. git commit → Execute with Bash (on approved)

Post-Implementation Verification (After All Tasks Complete)

Resolve the Work Plan's readable Design Doc, or the Work Plan itself when no Design Doc exists; missing input blocks verification.

Emit these Agent calls in one assistant message, then await both:

• code-verifier (subagent_type: "dev-workflows:code-verifier") → resolved doc_type, document_path, and code_paths from git diff --name-only main...HEAD
• security-reviewer (subagent_type: "dev-workflows:security-reviewer") → the same typed governingDocuments and implementationFiles

Apply subagents-orchestration-guide's Post-Implementation Verification pass/fail and fix/re-run rules. Present the unified report; proceed to Final Cleanup after both pass.

Final Cleanup

Before the completion report, delete the implementation task files this recipe consumed. Their work is committed; docs/plans/ is ephemeral working state and is not retained between recipe runs:

• Delete every file matching docs/plans/tasks/{plan-name}-task-*.md (the {plan-name} derived from the work plan path used in this run)
• Delete every file matching docs/plans/tasks/{plan-name}-phase*-completion.md (the per-phase completion files generated by task-decomposer)
• Delete the corresponding docs/plans/tasks/_overview-{plan-name}.md if present
• Preserve the work plan itself (docs/plans/{plan-name}.md) — the user decides whether to delete it after final review

If task files cannot be deleted (filesystem error), report the failure but do not block the completion report.

Test Information Communication

After acceptance-test-generator execution, when invoking work-planner (subagent_type: "dev-workflows:work-planner"), communicate:

• Generated integration test file path (from generatedFiles.integration)
• Generated fixture-e2e test file path or null (from generatedFiles.fixtureE2e)
• Generated service-integration-e2e test file path or null (from generatedFiles.serviceE2e)
• Per-lane E2E absence reason (from e2eAbsenceReason.fixtureE2e and e2eAbsenceReason.serviceE2e, when each lane is null)
• Explicit note: integration tests are created simultaneously with implementation, fixture-e2e tests are created alongside the UI feature phase, service-integration-e2e tests are executed only in the final phase

Execution Method

All work is executed through sub-agents. Sub-agent selection follows subagents-orchestration-guide skill.