| name | plan |
| description | Generate an implementation plan from a locked per-task spec, gated by human approval before any code is written. Use after the spec exists and before implementation. Triggers on /plan, "make a plan", "draft a plan for this spec". |
/plan
Invoke the planner subagent against the current per-task spec to produce an implementation plan for human approval (gate 1 of the workflow loop).
When to use
- A per-task spec exists (in
docs/specs/<feature>/<NN>-<slug>.md or similar).
- No code has been written yet for this slice.
- You want to gate the agent before implementation.
Steps
- Identify the spec file. Ask the user if ambiguous.
- Read the spec, AGENTS.md / CLAUDE.md, and the canonical example the spec cites.
- Spawn the
planner subagent with the spec path. Pass it:
- The spec file path
- The AGENTS.md / CLAUDE.md path
- The canonical example path (extracted from spec's Constraints)
- The planner produces a plan to stdout (or writes to
<spec>.plan.md).
- STOP. Show the plan to the user. Do not proceed to implementation.
Output
A plan with the structure defined in agentic-playbook/subagents/planner.md:
- Approach (prose)
- Files to add/modify
- Canonical pattern being mimicked
- What this slice will NOT do
- Abstractions introduced (with justification)
- Risks
- Done check
Hard rules
- The plan is the deliverable. Do not start implementing.
- If the user pushes back on the plan, revise the plan and re-show. Do not jump to code.
- If the spec is broken (missing constraints, ambiguous, contradictory), stop and ask before drafting a plan.
Next step after plan is approved
Use /implement.