Found a bypass, a false negative, or any security issue? That is the most valuable contribution this project can receive. Report it:

• Preferred: open a veto-story issue (public is fine for deny-list gaps — the bypass suite already prints its known gaps).
• For anything sensitive (e.g. a vulnerability in the hook itself, not just a pattern gap): email bgml@bgml.ai. You'll get a reply within 72 hours.

Reported gaps get fixed and credited in the CHANGELOG.

gate.cat's own doctrine applies to itself: the gate is certain only about what it blocks. An action it does not match is unchecked, not safe. A new unmatched-but-dangerous shape is an expected finding, not an embarrassment — please send it.

• In scope: deny-list gaps, analyzer false negatives, fail-open behaviors (anything where an error path ALLOWS instead of blocking), hook-integration bypasses, false-positive classes.
• Known, documented gaps (printed by bypass_suite): a Unicode-homoglyph binary name (ｒm), a printf-hex-assembled rm piped to a shell, and the runtime-assembled $'\x72m' (a regex-wall gap the delete-analyzer still catches) — improvements welcome, reports not needed. Note: base64/curl|sh/wget|python and runtime shutil.rmtree are NOT gaps — the first three are blocked by ENCODED_EXEC, the last is surfaced as a warn by RUNTIME_DELETE.

Only the latest release on PyPI receives fixes.

No bounty program (solo, pre-revenue). If you plan to publish a working bypass, a 14-day heads-up before publication is appreciated — most pattern fixes ship in days.