refresh-index

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

• Third-party content exposure detected (medium risk: 0.30). SKILL.md’s runtime workflow refreshes enabled GitHub repos by running npm run preindex and then reads the resulting per-repo JSON from data/skill-index/*.json, where preindex necessarily ingests repository contents cloned from upstream (outsider-authored), making free text from those repos LLM-readable during ingestion.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). This skill explicitly clones upstream repositories from github.com at runtime (see "Network access to github.com" and examples like github:anthropics/skills) and invokes the ingester via npm run preindex (which relies on asm eval), so fetched repo content can directly influence agent data or trigger remote code evaluation.