| name | initiative-planning |
| description | Plan a large or ambiguous engineering initiative before implementation. Produces intent, discovery, plan, chunk map, and chunk contracts. Use before writing code for non-trivial work. |
Initiative Planning
Use this skill when the user describes a large, ambiguous, architectural, or multi-PR task.
Mode
Start read-only. Do not edit application code.
Inputs
- Initiative folder path, or create one under
.agent-loop/initiatives/.
- Human goal and constraints.
- Project policies under
.agent-loop/policies/.
Required output artifacts
Create or update:
INTENT.md
DISCOVERY.md
PLAN.md
CHUNK_MAP.md
STATUS.md
RISKS.md
DECISIONS.md
chunks/<ID>-01.md ... chunks/<ID>-N.md
Process
- Restate the human goal.
- Identify what is known, unknown, and risky.
- Explore the repo read-only.
- Write
DISCOVERY.md with concrete files, modules, tests, and risks.
- Write
INTENT.md with the why, success state, non-goals, and human decisions needed.
- Write
PLAN.md with proposed approach, alternatives rejected, boundaries preserved, and verification strategy.
- Split the plan into reviewable PR-sized chunks in
CHUNK_MAP.md.
- Create one chunk contract per chunk.
- Mark the first chunk as proposed, not active, until human approval.
- Stop. Do not implement.
Chunking rules
- Each chunk should map to one PR.
- Each chunk must have allowed files, not-allowed changes, acceptance criteria, risk class, reviewer set, and human review focus.
- L1 chunks must be small enough for careful human review.
- If a chunk is too large, split it.
Output format
End with:
- Initiative summary
- Proposed chunk list
- Main risks
- Human decisions needed
- Recommended first chunk
- Explicit stop: "Planning complete. Awaiting human approval before implementation."