| name | plan-feature |
| argument-hint | feature description |
| description | This skill should be used when the user asks to
"plan feature", "design feature",
or wants to plan and design a new feature before implementation.
|
Feature Implementation Planning
Description: $ARGUMENTS
Preparation
- Read
.claude/rules/architecture-conventions.md -- confirm current DO/DON'T rules
- Read
.claude/rules/project-status.md -- confirm work currently in progress
- Read
.claude/rules/project-overview.md -- confirm tech stack and structure
- Identify current domain list: use Glob pattern
src/*/ and exclude _core, _apps prefixes
Default Flow Position
- Steps:
framing (Phase 0) + approach options (Phase 1) + plan (Phases 2~4)
- Routes after: STOP at the approved Execution Packet. Execution is a separate, explicit step — invoke
/execute-plan, which advances the ledger to executing (clearing the plan→execute block) and routes to the implement skills (/new-domain, /add-api, /add-cross-domain, etc.) internally. To run a single implement skill or self-evident follow-up directly, prefix the next prompt with a [trivial]/[hotfix] token (the block honours plan-waiver tokens) — invoking an implement skill while the stage is still planned without a token is hard-blocked. Never auto-continue from planning into implementation in the same turn (ADR 054)
- Recursion guard: do not invoke
/plan-feature recursively. Implement skills must not call /plan-feature (planning happens before implement)
Procedure Overview
- Requirements Interview — 3-5 questions from 5 categories (Phase 0)
- Approach Options — propose 2-3 candidates with trade-offs, recommend one (Phase 1)
- Architecture Impact Analysis — layer, domain, DTO, cross-domain (Phase 2)
- Security Checkpoint — 6-item assessment matrix (Phase 3)
- Task Breakdown — skill mapping, supervision levels, execution order (Phase 4)
- Execution Packet — include Goal, Scope, Success Criteria, Selected Approach,
Architecture Impact, Task List, Verification Gates, and Review Gates.
- Work-ledger update — after task breakdown is confirmed, record goal/scope/plan to
.agents/state/current-work.json via:
from work_ledger import update_goal_scope_plan, update_workflow_state
update_goal_scope_plan(goal="<one-line goal>", scope="<domains/files>", plan="<task list>", updated_by="skill:plan-feature")
update_workflow_state(stage="planned", plan_ref="<issue, PR, or plan file>", tasks=[{"id": "1", "title": "<task>", "status": "pending"}], updated_by="skill:plan-feature")
Read docs/ai/shared/skills/plan-feature.md for detailed steps and output format.
Also refer to docs/ai/shared/planning-checklists.md for question bank and templates.
For complex, architecture-changing, governor-changing, or multi-task work, hand
the approved Execution Packet to /execute-plan as a separate step — do not
implement within /plan-feature. Stopping at the packet is mandatory (ADR 054).
