stn-skills

Supports Claude Code, Cursor, and Copilot CLI.

---

Why stn-skills

AI-assisted coding without structure produces predictable failures:

• Plans abandoned halfway — without task decomposition and checkpoints, multi-step implementations drift from the original intent
• No verification — completion claims backed by nothing but the agent's word. Research measures 20–27% quality degradation without per-step feedback
• Deprecated code persists — nobody verified the old APIs were actually replaced
• Planning skipped entirely — yet structured decomposition yields 12–18% higher correctness

stn-skills closes these gaps with evidence-backed structure at every stage. Every design decision is explored through multiple lenses before commitment. Every implementation step is independently verified by separate reviewer agents. Every completion claim is backed by fresh evidence.

---

The Pipeline

graph LR
    A["Brainstorming<br/><i>6 phases, 4 gates</i>"] -->|design spec| V1{"Handoff<br/>Validator"}
    V1 -->|validated| B["Plan Writing<br/><i>6 phases, 4 gates</i>"]
    B -->|implementation plan| V2{"Handoff<br/>Validator"}
    V2 -->|validated| C["Plan Execution<br/><i>7 phases, 3 gates</i>"]
    C -->|completion report| D(("Verified<br/>Feature"))
    E["Codebase Audit<br/><i>13 domains, 3 gates</i>"] -->|"complex findings<br/>[PIPELINE]"| A

    style A fill:#2563eb,stroke:#1d4ed8,color:#fff
    style V1 fill:#d97706,stroke:#b45309,color:#fff
    style B fill:#7c3aed,stroke:#6d28d9,color:#fff
    style V2 fill:#d97706,stroke:#b45309,color:#fff
    style C fill:#059669,stroke:#047857,color:#fff
    style D fill:#16a34a,stroke:#15803d,color:#fff
    style E fill:#dc2626,stroke:#b91c1c,color:#fff

Each skill works independently or as part of the pipeline. The session-init skill auto-loads at session start and routes to the correct skill based on pipeline state. Use /stn-skills:build-feature for the full pipeline, or invoke each skill separately. Complex audit findings can be escalated to the brainstorming → planning → execution pipeline for structured remediation.

---

Available Skills

Skill	Invoke	Description	Typical Duration
Build Feature	stn-skills:build-feature	End-to-end pipeline: brainstorming → plan-writing → plan-execution in one workflow.	30–90 min
Brainstorming	stn-skills:brainstorming	Multi-lens design exploration with adversarial review. Transforms vague requests into validated design specs.	5–25 min
Plan Writing	stn-skills:plan-writing	DAG-based task decomposition with zero placeholders. Every step has complete code, verification, and rollback.	5–35 min
Handoff Validator	stn-skills:pipeline-handoff-validator	Validates design specs and plans at pipeline boundaries before the next phase consumes them.	1–3 min
Plan Execution	stn-skills:plan-execution	Checkpoint-verified execution with drift detection, 3-stage review, circuit breakers, and fidelity scoring.	~3 min/task
Codebase Audit	stn-skills:codebase-audit	13-domain evidence-based repository audit with confidence scoring, optional auto-fix, and pipeline escalation for complex findings.	15–45 min
Quality Bootstrap	stn-skills:codebase-quality-bootstrap	Generates production-grade CLAUDE.md and hooks aligned with all 13 audit domains.	5–15 min
Auto-Discovery	stn-skills:session-init	Session-start auto-loading with pipeline-state awareness. Routes to the correct skill.	auto

---

Install