Audit: LLM tooling¶
What we check¶
Two rules. One is measured; the other is a judgment call and is not.
AGENTS.md and ARCHITECTURE.md¶
Measured by llm-tooling. Every project should have an
AGENTS.md and an ARCHITECTURE.md, and the check fails on either
being missing from the repository root. It reads only whether they
exist; their shape is
llm-doc-structure's to measure.
Skills for repetitive operations¶
Reviewed, not measured. Repetitive operations should be covered by a Claude skill where one helps. Things that are likely to need a skill include remembering to write unit or functional tests, updating documentation for user-visible changes, and so forth. Whether an operation is repetitive enough to earn one is a judgment no check makes, so a repository with no skills passes this criterion.
Template¶
No template -- these files are project-specific and must be written with knowledge of the project's architecture and workflows.
Projects¶
Per-project compliance for this criterion is regenerated on every run of the consistency audit: see the compliance page.