grilling

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). The skill explicitly instructs automated lookup of environment/filesystem facts and pushes session records to an external repo URL taken from the DECISION_MEMORY_URL environment variable, creating a clear avenue for unauthorized data exfiltration of sensitive environment or session-local information.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly shallow-clones the repository given by the environment variable DECISION_MEMORY_URL at runtime and injects that repo's preferences.md/agent-instructions to drive slot 1 and recording behavior, so an externally supplied repo URL (DECISION_MEMORY_URL) directly controls prompts and agent behavior.