release-notes

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

• Third-party content exposure detected (high risk: 0.70). The skill's extract-pr-metadata flow includes an explicit --fetch-gh option that uses the gh CLI to fetch GitHub PR details (public, user-generated content) as documented in SKILL.md "PR Metadata Extraction" and in lib/extract-pr-metadata.cjs, and those PR bodies/labels are parsed/enriched and used by downstream steps (categorization, summaries, breaking detection) which can materially influence decisions.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill calls the Anthropic/Claude API at api.anthropic.com (POST https://api.anthropic.com/v1/messages) at runtime via lib/transform-llm.cjs to send prompts and receive model-generated content that directly controls the transformed release-note output, making it a required external runtime dependency.