AI covenant

AI covenant

AI_COVENANT.md says that everything a contributor submits is theirs, whatever tools helped make it, and that they must understand, verify and defend it.¹ Its note for this project says that the repository models someone else’s framework, the CIS Controls. It names two failure modes that are worse here than elsewhere:

• a rule or element a model produced, presented as what the CIS Controls say;
• CIS’s own text, which the repository may not carry.

Three rules follow:

• an element under src/cis_controls/schema/ says what the CIS Controls say, in the project’s own words, with the part of CIS’s guide cited in the pull request;
• no CIS text goes into the schema, the examples, the docs or the bundle, copied or closely paraphrased, and a model’s text is checked for CIS’s wording before it is submitted;
• everything the schema defines is in the cis_controls: namespace, with CIS’s numbers appearing only as values.

The covenant forbids posting AI output as an independent voice in issues or discussions. It sets four rules for the scheduled librarian:

• a bot identity or the maintainer’s, never both;
• every change as a pull request a person approves;
• a read-only job that hands a patch to a separate job that runs no agent;
• a recorded verdict only where a person said it.

AI co-authorship trailers in commits are discouraged.

Footnotes

1. AI_COVENANT.md, A note specific to this project and Repository-Owned Agent Automation ↩