The rules

What the PR does	Severity
Commits a secret; adds bidi/invisible-tag characters; pull_request_target plus a checkout of the PR's code; a dependency the supply-chain guard refuses; prompt injection in an agent-instruction file (AGENTS.md, CLAUDE.md, .cursorrules ...)	CRITICAL → BLOCK
Deletes, skips, empties, short-circuits or moves a test out of the tree; replaces exact assertions with weak ones; deselects tests in runner config; removes an auth decorator or inline access check	HIGH → REVIEW
Changes a CI pipeline, CODEOWNERS or an agent's permission settings; workflow script injection; prints a secret; write-all token; compiled binary; zero-width characters; text aimed at an AI reviewer	HIGH → REVIEW
Unpinned third-party action; unvetted dependency; swallowed exceptions; shell=True/eval; permissive CORS; debug on; visible injection phrase in a doc	MEDIUM (reported, still PASS)

Verdict: any CRITICAL → BLOCK; else any HIGH → REVIEW; else PASS. The Action fails the check on BLOCK by default (FAIL_ON=REVIEW to fail on HIGH too).

FIRME AI CORPORATION · CAIN-42 is a Byzantine trust fabric. Numbers marked "our own check" were run by us, not by an independent party. Nothing on these pages is a certification.