Create phased implementation plans from research artifacts. Use after /epic:explore completes to design atomic, verifiable implementation phases. Triggers on /epic:plan or when user wants to create an implementation plan, design phases, or prepare for coding based on completed research.
Limited to specific tools
Additional assets for this skill
This skill is limited to using the following tools:
references/plan-template.mdname: epic-plan description: Create phased implementation plans from research artifacts. Use after /epic:explore completes to design atomic, verifiable implementation phases. Triggers on /epic:plan or when user wants to create an implementation plan, design phases, or prepare for coding based on completed research. allowed-tools:
Planning phase of the Explore-Plan-Implement workflow. Create detailed, phased implementation plans based solely on research findings.
1. Locate active workflow: .claude/workflows/*/state.md
2. Verify Explore phase is complete
3. Load research artifacts (codebase-research.md, docs-research.md)
4. Design atomic, verifiable phases
5. Write plan to plan.md
6. Update state.md
7. Present summary for human review
If prerequisites not met:
⚠️ Cannot create plan - Explore phase incomplete
Run `/epic:explore [feature]` first.
Each phase must be:
## Phase N: [Descriptive Name]
### Changes
| File | Action | Description |
|------|--------|-------------|
| `path/to/file.ts:45-67` | Modify | [what changes] |
| `path/to/new.ts` | Create | [what it does] |
### Implementation Details
[Specific code changes, following patterns from research]
### Verification
**Automated:**
- [ ] `npm test` - [what it verifies]
- [ ] `npm run lint` - passes
- [ ] `npm run typecheck` - passes
**Manual:**
- [ ] [Specific check requiring human verification]
| Check | Requirement |
|---|---|
| Research-based | All decisions reference research findings |
| Specific | File paths include line numbers |
| Verifiable | Every phase has automated + manual checks |
| Complete | Rollback plan and success criteria defined |
| Right-sized | No phase too large for one session |
See references/plan-template.md for full structure.
After writing the plan, invoke the plan-validator agent to validate it before presenting to user.
Task tool:
subagent_type: epic:plan-validator
run_in_background: false
prompt: |
Validate the implementation plan at: .claude/workflows/[slug]/plan.md
Research artifacts: .claude/workflows/[slug]/codebase-research.md
Write validation report to: .claude/workflows/[slug]/plan-validation.md
IMPORTANT: Always launch a fresh agent. Never use the resume parameter.
| Verdict | Action |
|---|---|
| PASS | Present plan to user for approval |
| PASS WITH CONCERNS | Present concerns, ask user to review |
| NEEDS REVISION | Fix issues, re-validate before presenting |
✓ Plan Phase Complete
Feature: [description]
Phases: [N] phases defined
Artifact: .claude/workflows/[slug]/plan.md
Validation: [PASS/PASS WITH CONCERNS]
- Completeness: ✓
- Feasibility: ✓
Phase Summary:
1. [Phase 1] - [files affected]
2. [Phase 2] - [files affected]
Next: Review plan, then run /epic:implement
Error in Research → 1000s of bad lines
Error in Plan → 100s of bad lines ← THIS PHASE
Error in Code → ~1 bad line
This is the second-highest leverage point. Human review here prevents cascading errors.
references/plan-template.md - Full implementation plan template