A governance framework for AI coding agents — structured as four layers borrowed from enterprise IT governance.
A ready-to-use set of skills and rules for AI-assisted development, structured as four governance layers:
| Layer | What it does | How it works |
|---|---|---|
| Policy | Rules the agent follows every session | A persistent instruction file that loads automatically at session start |
| Procedures | Step-by-step workflows for defined tasks | Skill documents you invoke when needed |
| Controls | Quality gates that catch mistakes | Checks that run before every commit |
| Segregation of Duties | Agent proposes, human approves | Built into every skill at key decision points |
Seven skills covering the full product development lifecycle:
| Skill | What it does |
|---|---|
| Requirements | Takes a vague idea → researches feasibility → creates a well-structured ticket with testable acceptance criteria |
| Sprint | Full ticket workflow: pre-flight → pickup → research & plan (with your approval) → implement → quality gate & commit → UAT → merge. Supports single-ticket and batch modes. |
| UAT & Triage | Structured testing report (what was verified, what you can test, what's blocked), then proper triage: bug, new feature, or docs gap? Bugs get fixed. Features go to backlog. No scope creep. |
| Design-First | Wireframe with real components → get a clickable prototype → lock the design → implement against it. The wireframe is the source of truth. |
| UX Design | UX framework for information-dense apps: progressive disclosure, persistent context, scannable structure, forgiving navigation, zero-surprise interactions |
| Session End | Clean wrap-up: commit, push, update handoff docs, update tickets. Never end with uncommitted work or stale documentation. |
| Retrospective | Reviews recent session friction against current skills and rules. Proposes specific improvements. You approve each change. |
The flow: Requirements → Design-First + UX Design → Sprint → UAT → Session End → Retro
Or: define what to build → design it → build it → test it → wrap up → improve how you work.
The simplest approach. Open any skill file in skills/, copy the content, and paste it into your AI coding tool at the start of a session. The skill documents are written as instructions any AI coding agent can follow.
Most AI coding tools support a persistent instruction file that loads every session. Copy policy/POLICY.md into whatever your tool uses for this. Copy skills into your tool's skill or prompt system if it has one, or keep them as paste-in templates.
process-over-prompts/
├── README.md
├── LICENSE
│
├── policy/
│ └── POLICY.md ← Global rules (adapt to your tool's config)
│
└── skills/
├── sprint/SKILL.md ← Full ticket workflow (single + batch)
├── requirements/SKILL.md ← Idea → structured ticket
├── uat/SKILL.md ← Testing report + triage
├── design-first/SKILL.md ← Wireframe before logic
├── ux-design/SKILL.md ← UX framework for complex apps
├── session-end/SKILL.md ← Clean session wrap-up
└── retro/SKILL.md ← Improve your own process
You don't invoke every skill every session. Most of the time you just work normally — your policy file protects you in the background.
When you want the full structured workflow, use the skill that fits:
- Working through a ticket? → Sprint
- New idea that needs scoping? → Requirements
- Finished building, need to test? → UAT
- Starting a new screen? → Design-First
- Done for the day? → Session End
- Want to improve your setup? → Retro
Think of it like a workshop: safety goggles are always on (policy). Power tools are on the wall (skills). You grab the one you need.
MIT — use it, adapt it, share it.