narrative-quality-auditor

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

• Third-party content exposure detected (high risk: 0.85). The skill’s runtime path explicitly ingests “live brand surfaces (User-provided or scraped via firecrawl.py)” and “proxy reads” from public web sources (e.g., tavily.py, gdelt.py, bluesky.py, wayback.py), which can include outsider-authored free text from arbitrary pages/URLs into the LLM context for scoring and veto checks.

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

• Potentially malicious external URL detected (high risk: 1.00). The skill's standalone-install fallback explicitly instructs fetching runtime reference files from https://raw.githubusercontent.com/aaron-he-zhu/aaron-marketing-skills/main/references/, which the auditor must "Read" before scoring — a runtime fetch of raw GitHub content that directly controls the agent's instructions and is a required dependency in that mode.