菜单
Auto-loaded at session start. Resumes active pipelines, routes tasks to stn-skills. Triggers: session start, /stn-skills:session-init.
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
| name | session-init |
| description | Auto-loaded at session start. Resumes active pipelines, routes tasks to stn-skills. Triggers: session start, /stn-skills:session-init. |
If you were dispatched as a subagent to execute a specific task, skip this skill entirely. Only the top-level orchestrator routes.
If CLAUDE.md contradicts a skill instruction, follow CLAUDE.md. The user is in control.
ACTIVE PIPELINES ARE RESUMED FIRST.
THE STATE FILE IS THE TRUTH.
If plan mode is active and a skill matches the user's intent, invoke it — the skill itself handles exploration, planning, and execution with proper structure.
If an Active Pipeline State section appears below, a pipeline is active.
No Active Pipeline State section? Use routing below.
Match by INTENT, not literal keywords. The user may write in any language. Classify what the user wants to accomplish, then match to the closest skill.
| Priority | Intent | Skill |
|---|---|---|
| 1 | Active pipeline state | Resume stn-skills:{active_skill} |
| 2 | Build, create, implement, or deliver a feature or significant change — any task touching multiple files that benefits from design-before-code | stn-skills:build-feature |
| 3 | Explore options, compare approaches, make an architectural or design decision | stn-skills:brainstorming |
| 4 | Decompose requirements into tasks, create an implementation plan | stn-skills:plan-writing |
| 5 | Execute an existing plan step by step | stn-skills:plan-execution |
| 6 | Audit, review, or assess repository health, security, or quality | stn-skills:codebase-audit |
| 7 | Set up or update project quality standards, generate CLAUDE.md | stn-skills:codebase-quality-bootstrap |
Multiple match? Use higher priority.
No match? Proceed without a skill. Only use skills when they genuinely add value.
Do NOT invoke when: simple question, single-file fix, no code changes, user says "skip".
brainstorming -> plan-writing -> plan-execution. Each outputs an artifact for the next. build-feature orchestrates all three. codebase-audit can initiate a pipeline when findings require design or multi-file remediation — it writes the pipeline state and hands off to brainstorming or plan-writing. State: .claude/stn-skills-pipeline-state.json.
Use when executing a written implementation plan with verification at every step. Checkpoint-based recovery, drift detection, three-stage review, circuit breakers, and execution fidelity scoring. Every completion claim requires fresh evidence. Triggers: "execute plan", "run this plan", "implement this plan".
Use before any feature design, approach exploration, or architectural decision. Required before plan-writing. Dispatches parallel subagents for multi-lens analysis, weighted scoring, and adversarial stress-testing. Produces validated design specs. CHAINS TO plan-writing on completion — MUST invoke via Skill tool. Triggers: "brainstorm", "design", "explore approaches", "how should we build".
Use when a design spec or requirements need decomposition into executable tasks, before writing any code. DAG-based task decomposition with zero placeholders, complete code in every step, adversarial verification, and rollback planning. CHAINS TO plan-execution on completion — MUST invoke via Skill tool. Triggers: "write a plan", "plan this", "break this down".
Use when auditing a repository for production readiness, security, or code health. Dispatches up to 13 specialized auditor subagents in parallel with file:line evidence for every finding. Technology-agnostic, works with any language and framework. Triggers: "audit", "review the repo", "code health", "security scan".
Use for end-to-end feature delivery from idea to verified code. Orchestrates brainstorming, plan-writing, and plan-execution with validated handoffs and 11 gates. No shortcuts, no skipped gates, no unverified claims. Triggers: "build feature", "build this", "implement end-to-end", "full pipeline".
Use when setting up or updating project quality standards. Generates tech-stack-specific CLAUDE.md and hooks aligned with all 13 audit domains. Preventive counterpart to codebase-audit — bootstrapped projects produce zero audit findings. Triggers: "bootstrap", "set up standards", "generate CLAUDE.md".