triple-check

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). The skill requires returning the exact command and the "smallest untouched decisive output excerpt" (verbatim), which forces the LLM to include unredacted command text and raw outputs that may contain API keys, tokens, or passwords.

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

• Attempt to modify system services in skill instructions detected (high risk: 0.80). The skill explicitly covers reviews "across ... systems" and instructs the agent to "apply every queued fix" and "run decisive verification" (including commands and working directory), which directs the agent to make potentially state-changing modifications and execute commands that could alter system files or require elevated privileges.