mirror of
https://github.com/magnus919/agent-skills.git
synced 2026-09-18 06:56:59 +03:00
Magnus HedemarkandGitHub
f73162d176
* feat: add GxP AI governance coverage * fix: complete GxP governance attribution and evaluation coverage * fix: complete GxP provenance references * fix: expose GxP routing and complete research provenance * chore: regenerate marketplace metadata * chore: regenerate marketplace metadata * chore: regenerate llms skill catalog
73 lines
3.8 KiB
Markdown
73 lines
3.8 KiB
Markdown
# ai-governance
|
|
|
|
Design and run an organization's AI governance system: principles, operating models and decision
|
|
rights, risk frameworks, lifecycle gates, and the fairness, transparency, privacy, security,
|
|
regulatory, and board-oversight controls that make AI accountable.
|
|
|
|
## Why Install This Skill
|
|
|
|
AI systems get deployed faster than the organizations that build them can decide who is
|
|
accountable for them, what risk they are allowed to carry, and what evidence should gate each
|
|
stage of their life. Without a working governance system, launches are rubber-stamped, risks are
|
|
surfaced too late, and "someone approved it" is rarely traceable to a named, evidence-backed
|
|
decision. This skill gives your agent a complete, field-tested framework for standing up and
|
|
operating AI governance — not a compliance checklist you copy, but a method you run.
|
|
|
|
After installing, your agent can stand up a governance program from scratch, tier AI use cases by
|
|
risk and prescribe the controls each one requires, review an LLM or agent system for governance
|
|
and safety gaps, map a regulation to a concrete compliance and control plan, score organizational
|
|
governance maturity and get a prioritized gap list, and prepare board-level reporting. It ships
|
|
dense references for each governance domain, including a GxP/data-integrity overlay, six fillable templates, and two executable scripts,
|
|
so the method turns into working artifacts instead of advice.
|
|
|
|
## What You Get
|
|
|
|
| Path | What it provides |
|
|
|---|---|
|
|
| `SKILL.md` | The router: triggers, what the skill owns vs. doesn't, and when to load each file |
|
|
| `references/` (12 files) | Dense, scannable guides: principles, operating model, risk frameworks, lifecycle, fairness, transparency, privacy, LLM/agent security, regulation, procurement/board oversight, source index, and a GxP/data-integrity overlay |
|
|
| `templates/` (6 files) | Fillable artifacts: governance charter, use-case intake, model risk assessment, model card, third-party due diligence, board report |
|
|
| `scripts/governance-maturity.py` | CLI that scores an organization's governance maturity from JSON answers and lists gaps |
|
|
| `scripts/use-case-risk-tier.py` | CLI that classifies an AI use case into a risk tier and its required controls |
|
|
| `evals/evals.json` | Output-quality cases used to grade the skill |
|
|
| `README.md` | This human-facing overview |
|
|
|
|
## Quick Start
|
|
|
|
The skill is pure methodology plus two stdlib-only Python CLIs — there is nothing to install or
|
|
configure.
|
|
|
|
Score governance maturity from an answers file:
|
|
|
|
```sh
|
|
python3 ai-governance/scripts/governance-maturity.py path/to/answers.json --json
|
|
```
|
|
|
|
Classify a use case's risk tier:
|
|
|
|
```sh
|
|
python3 ai-governance/scripts/use-case-risk-tier.py path/to/use_case.json --json
|
|
```
|
|
|
|
Both scripts print a single JSON object; add `--dry-run` to preview without writing anything.
|
|
Example input shapes are documented in each script's `--help`.
|
|
|
|
## Triggers
|
|
|
|
Load this skill when you or your agent need to:
|
|
|
|
- Stand up or mature an AI governance program, or design the operating model and decision rights.
|
|
- Tier an AI use case by risk and decide which controls it needs before it ships.
|
|
- Review an LLM or agent system (e.g., an internal RAG copilot) for governance and safety gaps.
|
|
- Map a current regulation to a compliance and control plan.
|
|
- Score organizational governance maturity and prioritize gaps.
|
|
- Prepare board-level AI governance reporting or run third-party/model due diligence.
|
|
- Govern AI used in a GxP context, including ALCOA+, data integrity, electronic records, validation/assurance, audit trails, or QMS interfaces.
|
|
|
|
## Requirements
|
|
|
|
- Nothing to install for the methodology or the templates.
|
|
- The two scripts need Python 3 (standard library only; no third-party packages).
|
|
- No API keys, accounts, or external services.
|
|
- Note: this skill provides governance guidance, not legal, financial, or security advice.
|