Start › Overview

Docs.

Start here

Pick the entry that matches what you’re trying to do.

AdoptQuickstart

Choose an audit path, install the skills, run the first audit, and review the generated artifacts.

first audit→ReadFull specification

29 clauses across 7 pillars. The normative document: every MUST, MUST when applicable, SHOULD, and MAY.

spec rev v0.1.2→AuditNo-install prompt

Drop a no-install audit prompt into a tool-using agent and produce stamped artifacts.

prompt path→ReferenceLevels overview

5 levels, L0-L4. Each level adds one gate on top of the previous.

level ladder→

At a glance

The spec is small on purpose. One page of normative text, one ladder of conformance levels, one canonical loop for verifying conformance.

Pillars7

Clauses29

Levels5L0-L4

Spec revv0.1.2

Docs by section

The sidebar, breadcrumbs, pagination, source links, search index, and this overview all derive from docs.config.json.

Section	Pages	Entries
Start	3	Overview · Quickstart · Adoption guide
Specification	6	Full specification · Pillars overview · Levels overview · Coverage map · Rule catalog · Threat model template
Audit	4	How the audit works · The profile · No-install prompt · A stamped audit
Skills	3	ai-contributor-audit-profile · ai-contributor-audit · ai-contributor-audit-fix
Project	7	Release Changelog · Contributing · Authors · Tooling · Security policy · Glossary · Published pages

The seven pillars

Each pillar groups clauses that share a concern. Names, ranges, icons, and descriptions come from the rule catalog projection.

§1–4

Engineering Foundation

The reproducible environment, static correctness, architecture boundaries, and pre-commit / CI gates that make any change reviewable.

pillar-I→

§5–9

Security

Secrets handling, dependency and CI/CD security, authorization boundaries, and threat modeling.

pillar-II→

§10–14

Quality & Reliability

Runtime validation, testing strategy, accessibility, failure handling and observability, and performance and reliability.

pillar-III→

§15–16

Release

Supply-chain transparency, branch protection, and release governance.

pillar-IV→

§17–20

AI Agents

How AI agents, shared skills, MCP servers, and delegated agents are governed in the repository.

pillar-V→

§21–22

AI Risk

AI-specific risks (prompt injection, untrusted input, capability scoping, allowlists, cost ceilings) and data protection for AI workflows.

pillar-VI→

§23–29

Oversight

Human approval, guardrail evidence, policy ownership, AI licensing and attribution, AI credential lifecycle, model/provider changes, and AI incident response.

pillar-VII→

Conformance levels

The level ladder describes how much of the workflow an AI tool participates in. Each level is cumulative.

Min. levelNameAutonomyWorkflowDefinition

L0Baseline Hygiene0%AI is not part of the contribution workflow yet. People may still use AI for personal help, but no AI tool reads the repository as workspace context, creates commits, or opens pull requests.The repository satisfies the baseline requirements in §1, §2, and §5: no secrets in version control, placeholder-only env examples when contributor-supplied or runtime environment variables exist, documented credential handling, a clean setup path, committed lockfile, pinned runtime, pinned package manager, and automated formatting. AI is not yet part of the contribution workflow. A repository at Level 0 publishes `conformance_level: 0`.

L1Hardened25%AI tools may read repository context, explain code, suggest commands, or help with review, but do not produce shippable changesLevel 0, plus the remaining unconditional `MUST` rows in Pillars 1–4 and the oversight MUSTs in §23, §24, and §25. A Level 1 repository is strong enough for AI tooling to read repository content. AI-specific MUSTs in Pillars 5–6 are not yet evaluated.

L2AI Assisted50%AI may produce code, docs, tests, configuration, or migrations that a human actively accepts. Extra risks such as tool use, external GitHub content, fetched URLs, retained context, provider routing, and dependency suggestions must be controlled.Level 1, plus AI instructions, AI data classification, AI-surface redaction, AI-generated-content licensing, and authorship tracking. Any triggered §18–§22 or §26 rows must also be satisfied. AI may assist humans, but every change passes through human acceptance.

L3AI Authored75%AI may complete delegated implementation tasks, open pull requests, or make material repository changes for human reviewLevel 2, plus the rows triggered by AI-authored work, shared skills, MCP servers, delegated agents, or other AI tooling that materially authors code. `AIC-prompt-audit-trail` always applies because Level 3 means AI materially authors code that ships. Humans still review before merge.

L4AI Autonomous100%AI may merge, release, deploy, schedule recurring changes, approve workflows, rotate settings, or otherwise act without per-change human approvalLevel 3, plus any autonomous-runner rows and every applicable `SHOULD`. Each `SHOULD` row must be `Fulfilled` or `Not relevant` with a documented reason. `⚠️ Warning` is not passing for Level 4. This level is for repositories where AI ships changes without human review for each change.

Full normative requirements per level are on the Levels overview page.

How to read these docs

Use the Quickstart if you want to run an audit now. Use the Adoption guide if you are changing a repository. Use the Rule catalog if you are building tooling. Use Published pages to inspect what this site publishes and where each page comes from.

Versioning and stability

The site renders the pinned specification revision. Audit artifacts pin the spec revision they were stamped against, and the validator cross-checks that stamp. See Releases for the version timeline.

Contributing

Issues, pull requests, and audits of real repositories are welcome. Start with Contributing and the Security policy.