classify-media-relevance

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.80). The skill instructs the model to emit the original "url" values in its JSON output and to base classification on the URL itself, so if an input URL contains sensitive query tokens or signed credentials the LLM will be required to reproduce them verbatim, creating an exfiltration risk.

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

• Third-party content exposure detected (high risk: 0.85). SKILL.md describes that during the parse pipeline/crawl-mode fetches the extractor reads full-page markdown from a fetched (outsider) page and then this skill classifies media, meaning outsider-authored free text (page markdown/alt/context) is ingested into the LLM context at runtime.