CRITICALCWE-15
PEI-005 Skill YAML Policy Injection
Policy Enforcement Integrity: Attacks that bypass policy enforcement via threshold fallbacks, config file injection, or parser error paths that fail open
Description
Attempts to inject permissive policy via a skill definition file that overrides agent-level restrictions
Remediation
Skill definitions must be validated against a schema that prohibits policy_override fields. Skill files must be signed or loaded from a trusted path only.