atlas-audit

Score the repository evidence behind your team's AI-coding maturity in two seconds.

npx @agentic-atlas/audit

atlas-audit measures the seven dimensions that a repository can support with evidence. The full Agentic Atlas maturity model contains 12 dimensions: seven repository dimensions plus five human dimensions assessed through interviews.

  ▲ atlas-audit · repository maturity evidence
  ──────────────────────────────────────────────────────────────
  ##  Context files          Strong evidence across context file...
  #-  Agents and skills      Skills / commands: 1 skill defined
  --  MCP integrations       No MCP configuration
  ...
  ──────────────────────────────────────────────────────────────
  Score: 71/100  Structured

Most teams run Claude Code on defaults. The gap between installed and configured is where the productivity hides. This tool shows what your repository can prove without uploading code or telemetry.

The seven repository dimensions

| Dimension | Evidence used | |---|---| | Context files | Root and module CLAUDE.md files, structure, quality | | Encoded standards | Concrete architecture and workflow rules | | Agents and skills | Reusable agents, skills, and commands | | Security policy | Permissions, secret gating, written security rules | | Hooks and CI | Hooks, formatter or linter config, CI pipelines | | MCP integrations | Versioned MCP server configuration | | Config ownership | Versioning and freshness relative to repository activity |

Each dimension rolls up granular technical checks. Use --json to receive both the seven maturity dimensions and their underlying evidence checks.

Usage

npx @agentic-atlas/audit
npx @agentic-atlas/audit path/to/repo
npx @agentic-atlas/audit --json
npx @agentic-atlas/audit --md report.md
npx @agentic-atlas/audit --ignore fixtures --ignore generated
npx @agentic-atlas/audit --version

Zero runtime dependencies. Nothing leaves your machine. It is a purely local file-system scan. Only shared .claude/settings.json configuration contributes to the score. Local settings and fixture directories are excluded from repository evidence.

JSON output

Schema version 2 includes:

• score and band for the seven repository dimensions
• dimensions with the canonical repository rollups
• checks with granular technical evidence
• coverage: "repository" to distinguish this score from the full audit

The bundled lib/schema-v2.json file is the machine-readable JSON Schema for this output.

GitHub Action

Keep repository evidence from regressing:

name: Claude Code maturity
on: [pull_request]
jobs:
  audit:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: agenticatlashq/atlas-audit@v1
        with:
          min-score: 50

The score and seven-dimension table appear in the job summary.

Scaffold the baseline

npx @agentic-atlas/audit fix
npx @agentic-atlas/audit fix --dry-run

The command creates only missing baseline files and never overwrites existing configuration. Fill in every bracket before committing the result.

What requires humans

A repository cannot prove adoption depth, review behavior, measurement quality, training depth, or budget ownership. Those five dimensions complete the 12-dimension maturity audit through structured interviews and evidence review.

Full audit and prioritized roadmap: agenticatlas.dev

MIT © Agentic Atlas. Not affiliated with or endorsed by Anthropic.