skill-trace-evaluation

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.80). The skill directs the AI to scan SKILL.md for "密钥/secrets/脚本" and requires each sub-item to include a cited "evidence" snippet in the generated report, which may cause the model to copy secret values verbatim into its output.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.75). Required workflow Step 2 (“AI 直接阅读正文”) ingests the target skill’s SKILL.md and references/ files from the provided skill directory into the LLM context for semantic calibration; if that skill content is authored by an outsider, it becomes outsider-authored free text (indirect prompt injection risk).