code-reviewer

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.90). The skill embeds a "Use-to-Evolve" Post-Invocation Hook that mandates running and logging usage and even autonomously applying micro-patches ("Do not skip it", "apply autonomously") which instructs the agent to change its behavior and persist data beyond the stated read-only code-review purpose—a hidden/deceptive control flow outside the skill's advertised scope.

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.80). The skill's Post-Invocation Hook explicitly logs "trigger_input": "<first 80 chars of user input>" (writing it into .skill-audit/usage.jsonl) and the scan rules call out hardcoded credentials (CWE-798), which means the agent is instructed to capture and may output user-provided secret substrings verbatim, creating an exfiltration risk.