manifest

You can automatically generate repository metadata for CLAUDE.md by using a tool that parses pyproject.toml and the file tree to derive package facts, dependencies, and CI workflows. This ensures the documentation block stays up-to-date without manual editing.

A living spec for repository facts is a machine-authored, machine-readable block in CLAUDE.md containing derived metadata from pyproject.toml. It prevents documentation drift by using CI checks that fail builds if the generated block diverges from the actual source facts.

To enforce CI checks that keep CLAUDE.md metadata in sync with source files, you configure a validation script that compares the machine-owned block against facts derived from pyproject.toml and the file tree. The CI pipeline fails the build if any drift is detected.

Yes, you need Python 3.11 or higher to parse pyproject.toml for repository documentation because the process relies on the built-in tomllib module to read the configuration file and manipulate text blocks in CLAUDE.md.

Yes, you can integrate dependency management and code documentation layout into a single automated spec. The process derives package metadata, dependencies, and file tree layout from pyproject.toml, splicing them into a unified machine-owned block within CLAUDE.md.

The main limitation of using a machine-authored block for repo metadata in CLAUDE.md is that it is strictly machine-owned, meaning manual edits to that specific block are overwritten. It also requires Python 3.11+ and a valid pyproject.toml to function correctly.