The one narrow fix round after /code-review high on E (three findings of
new classes; the between-rounds findings are known issues, decision 226):
- threat-model.mjs isOwnSecurityFinding: the engines' key in rule_key is
never Voyage's own, also when rule_keys names a key of its own (keysOf
reads rule_keys alone when it is a list, so an entry with rule_key
SECURITY_REVIEW_VULNERABILITY and rule_keys [SECURITY_RISK] could be
backed, against decision 218).
- plan-gate.mjs: --brief without --record is a usage error (exit 2); the
bare gate and --accept-open read the verdict's brief_path, and a --brief
there was silently ignored.
- verdict-gate.mjs: the known_outside check reads the threat model from the
answers' repo (--repo, else the review file's own work tree); with no
work tree nothing is backed. A review outside any work tree, gated from
another repo, was backed by that repo's brief (ALLOW).
Each with a test seen failing on an assertion first.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Kjell Tore Guttormsen
d402b0088b