changelog

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

• Third-party content exposure detected (medium risk: 0.30). Skill.md ingests Claude Code changelog text either from user-pasted conversation content or from webfetching the first-party changelog URL, so outsider-authored free text can be directly provided (pasted) into the runtime ingest phase.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly performs WebFetch of the changelog at https://code.claude.com/docs/en/changelog (appearing as code.claude.com/docs/en/changelog in SKILL.md) at runtime and uses that fetched content to drive the fetch/diff/apply pipeline and subagent research, meaning remote content directly influences agent instructions and required behavior.