version-check

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

• Third-party content exposure detected (medium risk: 0.30). The required workflow ingests outsider-authored free text from GitHub issue/PR titles/bodies via gh api search/issues and from Reddit threads via reddit/search.json, both at runtime before recommending versions.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill runtime uses curl to fetch remote markdown that directly informs the agent's recommendations (CHANGELOG entries), specifically https://raw.githubusercontent.com/anthropics/claude-code/main/CHANGELOG.md which is pulled and parsed at runtime.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 0.80). The skill explicitly describes a method to bypass Reddit's automation blocks (a security circumvention via a DuckDuckGo/Playwright "hop") and recommends system-modifying actions (global npm install and disabling the auto-updater), so it encourages bypassing protections and changing the host state.