Detection rules

Every Lucin detection rule: what it catches in an AI agent, why it matters, and how to fix it. 41 documented rules with severity and OWASP Agentic mapping.

Each rule below has a stable ID that appears in every output format, so a finding in CI is the same object as a finding in your terminal. Severity is bounded by evidence: a finding with no witness and no source line is capped below HIGH, because a reader cannot verify it.

Coverage, honestly

41 rules are documented here. That is not the same as 41 rules running — lucin scan --list-rules prints the set that actually fires, and this page is generated from that same registry, so the two cannot drift. Documented but deliberately not registered: AG-013, AG-013-T3, AG-PATH-TRAVERSAL. A detector is registered only once its false-positive behaviour against the benign corpus is understood. Precision is chosen over recall, and what that costs is published on limits rather than hidden. What the rule set as a whole does and does not catch is measured on benchmarks and limits.

pip install lucin
lucin scan .