maxkle1nz/kognit1v-intelligence-systems

Welcome to kognit1v-intelligence-systems. Use #help for questions, #ideas for proposals, and #showcase for what you build.

test

Code discussion: scripts/validate-kognit1v.mjs · L48-L55 The validator refuses to ship secrets by pattern, not by filename. The last pattern is the one that catches real accidents: a live SOMETHING_KEY=value assignment pasted into a doc or an example. The named-provider prefixes (sk-, sk-ant-, AIza) catch the classic cases, but env_assignment_secret is deliberately broad — better a false positive I have to clean up than a token in a public repo.

Code discussion: contracts/proof-contract.schema.json · L6 This required list is the doctrine in schema form: a proof contract is invalid unless it declares negative_proofs (what was checked and shown NOT to happen) and non_claims (what this proof deliberately does not assert). Positive assertions alone read like marketing. Forcing every proof to carry its own boundaries is what keeps agent-generated claims honest.

Code discussion: scripts/validate-kognit1v.mjs · L167-L181 Validation doesn't just pass or fail — it emits a proof object with counts, the failure list and a digest of the required-path set, and --write-proof persists it to proofs/validation/latest.json. The point: "the repo validates" becomes a claim you can diff, not a green checkmark you have to trust. The digest ties the proof to the exact contract it validated against.