| title | Plan Mode |
|---|
Use wavemill plan when work is still too large or ambiguous for wavemill mill. It decomposes a broad epic or initiative into well-scoped sub-issues that can be milled later.
- Selects an epic from your Linear backlog.
- Optionally researches comparable products and patterns.
- Decomposes the epic into 3–10 independently executable sub-issues.
- Creates a milestone and all sub-issues in Linear with priority, estimates, and dependency links.
# interactive epic selection + decomposition
wavemill plan
# or use the tool directly
npx tsx tools/plan-initiative.ts list # rank initiatives by size
npx tsx tools/plan-initiative.ts decompose # decompose + create issuesPick an initiative or large issue from your Linear backlog. Context is saved to epics/<epic-name>/selected-task.json.
Pass --research to have Claude study 2–3 comparable products and extract patterns and anti-patterns before planning. Output goes to epics/<epic-name>/research-summary.md.
Claude breaks the epic into sub-issues organized by milestone:
- Proof of Concept — validate core assumptions
- MVP — minimum shippable feature set
- V1 Launch — polish, edge cases, observability
- Long-term — extensions, optimizations
Each sub-issue includes a title, user story, description, priority (P0–P3), and dependency references.
The decomposition plan is pushed to Linear:
- A milestone is created (or matched to an existing one).
- Sub-issues are created with enhanced descriptions that reference the parent epic and relevant files.
- Dependency relations (
blocks) are created between issues so the backlog reflects execution order.
| File | Purpose |
|---|---|
tools/plan-initiative.ts |
List and decompose initiatives |
shared/lib/wavemill-plan.sh |
Plan workflow orchestrator called by wavemill plan |
tools/prompts/initiative-planner.md |
System prompt for decomposition |
tools/prompts/research-phase.md |
Research phase prompt template |
shared/lib/linear.js |
Linear API helpers (create issues, milestones, relations) |
- Feature Workflow — guided single-issue execution with plan → implement → validate gates
- Mill Mode — autonomous parallel backlog processing
- Expand Mode — batch expand issues into task packets
- Eval Mode — evaluate LLM performance on workflows
- Labels — how labels control parallel task scheduling
- Troubleshooting — common issues and fixes