Feature Implement Skill

Purpose

Operator Context

Hardcoded Behaviors (Always Apply)

• CLAUDE.md Compliance: Read and follow repository CLAUDE.md
• Plan Required: CANNOT implement without a plan in

  .feature/state/plan/

• State Management via Script: All state operations through

  python3 ~/.claude/scripts/feature-state.py

• Domain Agent Dispatch: Every task dispatched to its assigned domain agent via Task tool
• Wave Order Enforcement: Complete all tasks in Wave N before starting Wave N+1
• Wave Checkpoint: Run relevant tests after each wave completes
• Deviation Handling: Tier 1-2 auto-fix, Tier 3 stops for user

Default Behaviors (ON unless disabled)

• Context Loading: Read L0, L1, plan artifact, and design artifact at prime
• Fresh Agent Per Task: Each task gets a clean agent dispatch (no context pollution)
• Spec Compliance Check: After each task, verify output matches plan specification
• Progress Reporting: Report after each task and wave completion

Optional Behaviors (OFF unless enabled)

• Parallel within wave: Dispatch parallel-safe tasks simultaneously
• Auto-fix Tier 2 deviations: Handle missing dependencies automatically

What This Skill CAN Do

• Dispatch tasks to domain agents (golang-general-engineer, typescript-frontend-engineer, etc.)
• Execute wave-ordered plans with dependency tracking
• Handle deviations with tiered escalation
• Run wave checkpoints (tests) between waves

What This Skill CANNOT Do

• Implement without a plan
• Override domain agent selection from plan
• Skip wave ordering
• Handle Tier 3 (architectural) deviations without user input

Instructions

Phase 0: PRIME

1. Verify feature state:
   bash

   python3 ~/.claude/scripts/feature-state.py status FEATURE

   Verify current phase is

   implement

   and

   plan

   is completed.
2. Load plan artifact from

   .feature/state/plan/

   .
3. Load design artifact from

   .feature/state/design/

   for reference.
4. Load L1 implement context:
   bash

   python3 ~/.claude/scripts/feature-state.py context-read FEATURE L1 --phase implement

5. Capture BASE_SHA:
   bash

   git rev-parse HEAD

Phase 1: EXECUTE (Wave Dispatch)

1. Check if task is parallel-safe and parallel mode is enabled
2. Dispatch to assigned domain agent via Task tool:

Agent(
  subagent_type="[agent-from-plan]",
  prompt="Implement the following task:\n\n[TASK DETAILS FROM PLAN]\n\nFiles: [FILES]\nOperations: [OPERATIONS]\nVerification: [VERIFICATION COMMAND]\n\nContext from design: [RELEVANT DESIGN CONTEXT]",
  description="Implement [task title]"
)

1. Verify task output matches plan specification

python3 ~/.claude/scripts/feature-state.py gate FEATURE implement.architectural-deviation

1. Run the project's test suite (or relevant subset)
2. If tests fail: identify which task caused the failure, route back to that agent
3. If tests pass: proceed to next wave

Wave N complete: X/Y tasks passed
  T1: [agent] ✓
  T2: [agent] ✓
  T3: [agent] ✗ (deviation: [description])

Phase 2: VALIDATE (Implementation Review)

• All planned files were created/modified
• All verification commands pass
• No unplanned files were modified (check

  git diff --name-only BASE_SHA..HEAD

  )
• No Tier 3 deviations unresolved

Phase 3: CHECKPOINT

1. Save implementation artifact (summary of what was built):
   bash

   echo "IMPL_SUMMARY" | python3 ~/.claude/scripts/feature-state.py checkpoint FEATURE implement

2. Record learnings — if this phase produced non-obvious insights, record them:
   bash

   python3 ~/.claude/scripts/learning-db.py record TOPIC KEY "VALUE" --category design

3. Advance:
   bash

   python3 ~/.claude/scripts/feature-state.py advance FEATURE

4. Suggest next step:

   Implementation complete. Run /feature-validate for quality gates.

Error Handling

Anti-Patterns

References

• Gate Enforcement
• Retro Loop
• State Conventions
• Subagent-Driven Development