clj-skill-eval

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 1.00). The prompt explicitly tells operators to bypass platform permission checks (use --dangerously-skip-permissions) to avoid "permission constraints," which is an instruction to override safety/authorization controls and is outside the legitimate scope of evaluating Clojure skills.

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). The content contains high-risk instructions to bypass agent permission controls (e.g., using --dangerously-skip-permissions) and to launch unsandboxed subagents for parallel execution, which explicitly facilitates unauthorized code execution, unrestricted file/network access, and potential data exfiltration even though no direct exfiltration payloads or obfuscated backdoors are present.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The prompt explicitly instructs using --dangerously-skip-permissions to bypass permission constraints (i.e., circumvent security mechanisms), which encourages compromising the host's protective controls even though it doesn't directly request sudo or system-file edits.