Anatomia: Spec-Driven Agent Harness for Claude Code and Codex

What changed

Anatomia has been released as a CLI and agent harness that integrates with Claude Code and Codex. This tool is designed to scan codebases, identify project stacks, conventions, and patterns, and then process every change through a five-agent pipeline. The pipeline stages include scope, plan, build, verify, and learn, with each step generating and saving artifacts such as scope definitions, specifications, contracts, build reports, and verification results. The system emphasizes an auditable proof chain for all development activities.

Key commands include ana scan for detecting project specifics, ana init for generating context and agent definitions, and ana run to initiate the development pipeline. The ana init commit command persists infrastructure configurations to Git, ensuring team consistency. The tool also provides commands for project health checks (ana doctor), pipeline state (ana work status), and proof chain analysis (ana proof health, ana proof audit).

Anatomia's scan command analyzes a project in approximately 10 seconds, identifying frameworks, databases, authentication methods, testing strategies, services, and coding conventions. For monorepos, it detects individual surfaces and their associated commands. The init command writes this intelligence into configuration files like ana.json for project settings and scan.json for structured scan data, along with markdown files (CLAUDE.md, AGENTS.md) for cross-tool project context. It also generates skill templates and stack-specific guidance.

The five-agent pipeline operates as follows:

• Think (ana run): Acts as a thinking partner, handling scope definition and providing advice.
• Plan (ana run plan): Architects the solution, producing a sealed contract.
• Build (ana run build): Implements the specification and generates code and tests.
• Verify (ana run verify): Acts as a fault-finder, reviewing the spec and code against the build report.
• Learn (ana run learn): Analyzes proofs across cycles to improve skills and system performance.

Each pipeline run contributes to a proof chain, which can be tracked using ana proof health to monitor metrics like first-pass verification rates and risk areas. Recurring patterns can be promoted into skill rules using proof promote.

Why it matters for builders

Anatomia provides builders with a structured and auditable framework for leveraging AI coding assistants like Claude Code and Codex. By formalizing the development process through a pipeline and generating an auditable proof chain, it enhances transparency and reliability in AI-generated code. This allows developers to better understand, debug, and trust the code produced by AI agents, fostering more effective collaboration and integration of AI into their workflows.

Practical impact

Developers can begin using Anatomia by installing the anatomia-cli globally via npm, requiring Node.js 22+. After installation, they can run ana init to generate project context and agent definitions, followed by ana run to start the AI-assisted development process. For teams, ana init commit can be used to share configurations. Builders can explore the ana scan command to quickly assess their existing codebase's compatibility and structure with the harness. The tool's output, including scan.json and markdown files, is designed to be compatible with other AI tools, offering flexibility.

Caveats and source limits

The provided source details the functionality and commands of Anatomia but does not include specific benchmark results comparing its performance against other agent harnesses. Information regarding pricing, specific model versions supported beyond Claude Code and Codex, or detailed performance metrics for the five-agent pipeline is not present. The source also mentions a "fresh release" but does not specify an exact release date for version 1.2.2, only a general publication timestamp for the repository metadata.