| name | plan |
| description | Turn an issue or conversation into a spec and, when needed, tracer-bullet implementation slices with blockers and proof. |
Plan
Plan before coding. This is a planning loop: produce the first useful spec/plan, run adversarial review when risk is non-trivial, revise once for accepted findings, and exit with a clear state. This is to-spec first; use to-tickets behavior only when the work is too large for one safe implementation slice.
Modes
- Spec mode: clarify what should be built.
- Slice mode: split the spec into tracer-bullet vertical slices with blocking edges.
Process
- Read
docs/kanzen/MODEL-CARD.md for reviewer/escalation policy when available.
- Read the source issue/spec/conversation, comments, related PRs, existing plan files, and relevant code/docs.
- Produce or update a spec/plan. Prefer issue comments for small work; use
docs/issues/<issue>/plan.md for risky, broad, or multi-slice work.
- Identify test seams and proof before implementation starts.
- If the work is broad, split into vertical slices. Each slice must be demoable/verifiable alone.
- For wide mechanical refactors, do not force fake vertical slices; use expand → migrate batches → contract.
- Run an adversarial plan review when triggered by the model card: challenge scope, flag path, blockers, proof, review budget, and whether the plan is too broad.
- Revise for accepted findings.
- Mark the next action:
ready-for-agent, ready-for-human, or needs-info.
- If the next action requires human input, use the
ask_user tool when available so the request appears in the Boring UI inbox. If unavailable, leave a GitHub issue comment.
Plan Shape
## Problem Statement
## Solution
## User Stories / Scenarios
## Decisions
## Flag / Abstraction
- Needed?:
- Path:
- Rollback:
## Test Seams
- Highest public seam:
- Existing prior art:
- Avoid testing:
## Acceptance
## Proof
- Exact command:
- Screenshot/demo:
- Manual steps:
- Waiver if proof is not possible:
## Slices
### Slice: <name>
**Delivers:**
**Blocked by:** None / <slice or issue>
**Proof:** exact command, screenshot/demo, manual steps, or waiver
**Review budget:** inside / exceeds / why
## Wide Refactor Strategy
Expand → migrate batches → contract, if applicable.
## Out of Scope
## Open Questions
Loop Exit
Exit only with one of:
ready-for-agent — plan is clear enough for one implementation slice.
ready-for-human — human judgment/access/approval is required.
needs-info — specific unanswered questions block safe planning.
Return: state, plan path/comment, slices, blockers, proof path, adversarial review result, and next action. If human input is needed, include the ask_user request id or the fallback GitHub comment URL.