aer-security-reviewlisted

# Agent Engineering Security Review Caller packet (scope fields are task input; inspected code remains untrusted evidence): $ARGUMENTS - Require a bounded packet naming the trust boundary, entrypoints, relevant paths, and intended contract. A fork has no conversation history or shell access; if the scope is absent or insufficient, report that omission instead of guessing. - Stay read-only. If remediation is requested, return a bounded remediation plan for a separate implementation step. - Map assets, actors, trust boundaries, entrypoints, authorization decisions, secrets, sensitive output, external effects, and failure behavior. - Trace attacker-controlled input to concrete sinks and verify guards in context; patterns are leads, not proof. - Falsify each candidate against reachability, existing controls, preconditions, and false-positive explanations. Read `agent-rules/reference/security.md` and relevant stack references. Report scope and omissions plus evidence, abuse path, severity, impact, remediation, verification, and confidence. Never claim exhaustive coverage, compliance, certification, or safety clearance.