stefancgillberg Epic Plan
Purpose
Transform the story list from /rai-epic-design into a sequenced implementation plan with milestones, parallel work streams, and progress tracking.
Mastery Levels (ShuHaRi)
- Shu: Follow all steps, create sequenced plan with walking skeleton + MVP milestones
- Ha: Adjust milestone granularity based on epic size, skip timeline for small epics
- Ri: Domain-specific sequencing patterns, team velocity models
Context
When to use: After /rai-epic-design has produced a scope document with stories. Before starting first story.
When to skip: Very small epics (2-3 stories) with obvious linear sequence. Emergency fixes.
Inputs: Epic scope document (work/epics/e{N}-{name}/scope.md), calibration data (if available).
Steps
Step 1: Review Epic Scope
Load and understand the epic scope document:
- Story list with sizes and dependencies
- Done criteria and risks
- Architectural decisions that affect sequencing
Step 2: Sequence & Rationalize
Order stories using these strategies (in priority order):
| Strategy | When | Rationale |
|---|---|---|
| Risk-first | High uncertainty features | Tackle unknowns early while energy is high and time for pivots remains |
| Walking skeleton | Architecture unproven | Build minimal E2E path first to prove architecture |
| Quick wins | Need momentum | Early deliverable value, validate tooling |
| Dependency-driven | Hard blockers exist | Unblock others on critical path |
For each story, document: position, rationale, dependencies (hard/soft/external), what it enables.
Identify parallel opportunities: Stories with no mutual dependencies, different codebase areas, or independent concerns can run concurrently.
Step 3: Define Milestones
Create 2-4 intermediate checkpoints:
| Milestone | Typical scope | Purpose |
|---|---|---|
| M1: Walking Skeleton | 1-3 stories (smallest E2E) | Prove architecture, enable integration |
| M2: Core MVP | 50-70% of stories | Demonstrate value, gather feedback |
| M3: Feature Complete | 100% planned stories | Ready for polish |
| M4: Epic Complete | Done criteria met | Ready for /rai-epic-close |
Per milestone: stories included, success criteria (verifiable), demo capability.
Integration checkpoint: For epics with multiple components (client/server, CLI/API, frontend/backend), schedule an E2E integration milestone before the final story. This checkpoint runs real infrastructure (docker compose, actual DB) and verifies cross-story contracts (auth headers, payload schemas, parameter limits). Unit tests with mocks cannot catch these mismatches — only real E2E validates the seams between stories.
Step 4: Setup Tracking
Add progress tracking to epic scope using templates/plan-section.md:
- Story sequence table with status/actual/velocity columns
- Milestone checklist with target dates
- Velocity assumptions from calibration data (if available)
Step 5: Update Scope Document
Append the implementation plan section to work/epics/e{N}-{name}/scope.md. Include:
- Story sequence with rationale
- Milestones with success criteria
- Parallel work streams (if any)
- Progress tracking table
- Sequencing risks (top 3)
Present plan to human for review before starting first story.
Output
| Item | Destination |
|---|---|
| Implementation plan | Appended to work/epics/e{N}-{name}/scope.md |
| Plan template | templates/plan-section.md |
| Next | /rai-story-design for first story in sequence |
Quality Checklist
- All stories sequenced with rationale
- Critical path identified
- At least 2 milestones (walking skeleton + MVP minimum)
- Dependencies verified — no cycles, blockers identified
- Parallel opportunities documented (or explained why sequential)
- Progress tracking in scope document
- NEVER over-plan — plans are hypotheses, not commitments
- NEVER sequence by size alone — use risk-first as default
- Multi-component epics include E2E integration checkpoint
References
- Plan template:
templates/plan-section.md - Previous:
/rai-epic-design(produces scope input) - Next:
/rai-story-designfor first story - Close:
/rai-epic-close - Calibration:
.raise/rai/memory/calibration.jsonl