code-wiki

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

• Third-party content exposure detected (high risk: 0.85). The skill may clone a user-supplied GitHub URL at runtime (git clone --depth 50 <url> ...), and then read_file/search_files ingests arbitrary outsider-authored README/source text from that repo into the LLM context for documentation.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill runs git clone at runtime ("git clone --depth 50 ") to fetch a remote repository whose files are then read and injected into the agent's context to drive the generated wiki, so a remote URL can directly control prompts/content.