predictive-maintenance

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.80). The prompt contains explicit telemetry instructions to silently check and append data to a user project directory (~/.claude/projects) for “self-evolution,” which is unrelated to the stated predictive‑maintenance audit and includes a “skip silently” behavior that is covert—constituting a hidden/deceptive instruction outside the skill’s purpose.

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.80). The skill requires scanning codebases and reporting specific file locations and findings (including configs and integration code) that may contain hardcoded API keys or tokens, and it does not instruct the agent to avoid or redact secret values, so the LLM could end up outputting secrets verbatim.