Documentation
Anatomia is an open-source pipeline that turns AI-generated code into verified code. It runs locally, installs as agents inside your repo, and produces a proof chain you can audit, replay, and trust — including the 163 proofs that built Anatomia itself.
The pipeline at a glance
Every shipped change passes through the same five stages. Each stage is a sealed agent with one job, specific inputs, and validated outputs. What's left behind is permanent — the artifacts between stages are your team's engineering memory, auditable and replayable.
scope.mdana→02PlanSpec the solution. Seal the contract.spec.md contract.yamlana-plan→03BuildImplement. Tag tests to contract.build_report.mdana-build→04VerifyIndependent fault-finding.verify_report.mdana-verify05LearnPromote findings to rules.skill filesana-learnWhat's in these docs
The Reference section is auto-generated from the CLI source on every commit. Everything else is hand-written and reviewed.
Your first run
From zero to a working ana scan on your repo. No login. Two minutes to scaffold.
Doing the work
Recipes for the moves you’ll make, in the order you’ll need them.
Files & commands
Every CLI command, every template, every artifact format. The source of truth.
Where to start
Pick the door that matches what you're trying to do today.
I want to see if this is real
Read one complete proof end to end — scope, contract, build report, verify findings, integrity seal. Five minutes; no install.
→ Open a real proofInstallingI want to run this on my repo
Scan in three seconds, no login. Init ships the pipeline into your repo. Requires Claude Code.
→ QuickstartOperatingI have it running and want depth
How sealed agents work, how to read a verify report, when to promote a finding, how to recover from a rejection cycle.
→ How it worksResources
From the proof chain
Anatomia is built with anatomia. 163 pipeline runs, 3,384 assertions, 928 findings. These six show what the system produces.