菜单
Guide a feature from requirements interview through architecture analysis, security check, and task decomposition.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Inspect drift between code, shared workflow references, and Claude, Codex, or Antigravity harness assets after architecture or workflow changes.
This skill should be used when the user asks to "sync guidelines", "document inspection", "check skill updates", "update project-dna", "sync patterns", "verify code-document consistency", or after architecture changes to verify Skills/AGENTS.md/CLAUDE.md match the actual code.
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.
Audit a domain or the full repository for architecture compliance using the shared architecture checklist and repository rules.
Review a pull request or local diff against the repository's shared Review Protocol (correctness, regression, stability, contract, architecture, security, governance).
Review a domain, file, or the full repository for OWASP-oriented security issues using the shared security checklist.
| name | plan-feature |
| description | Guide a feature from requirements interview through architecture analysis, security check, and task decomposition. |
| metadata | {"short-description":"Feature implementation planning"} |
framing (Phase 0) + approach options (Phase 1) + plan (Phases 2~4)$execute-plan, which advances the ledger to executing and routes to the implement skills internally. To run a single implement skill directly, use a [trivial]/[hotfix] token. Never auto-continue from planning into implementation in the same turn (ADR 054; Codex surfaces the drift as a Stop-time advisory, Claude hard-blocks)/plan-feature recursively. Implement skills must not call /plan-feature (planning happens before implement)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 and workflow state via
from work_ledger import update_goal_scope_plan, update_workflow_state; update_goal_scope_plan(goal=..., scope=..., plan=..., updated_by="skill:plan-feature"); update_workflow_state(stage="planned", plan_ref=..., tasks=..., updated_by="skill:plan-feature")
Read AGENTS.md and docs/ai/shared/skills/plan-feature.md for the full procedure.
Read docs/ai/shared/planning-checklists.md for question bank and templates.
Interview the user on requirements (data model, business rules, integrations).
Propose 2-3 approach options with trade-offs and recommend one.
Analyze architecture impact, run security checkpoint, break into tasks.
Present the implementation plan in the standard output format, including the Execution Packet.
After approval, write the ledger (stage="planned") and stop — hand
complex, architecture-changing, governor-changing, or multi-task work to
$execute-plan as a separate step. Do not implement within $plan-feature
(ADR 054; on Codex the plan→execute drift surfaces as a Stop-time advisory).