forge

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

• Third-party content exposure detected (high risk: 0.75). Forge’s runtime LLM context can ingest outsider-authored free text when {from_file} points to a GitHub issue (gh issue view <n> --json title,body,labels,comments) or a fetched web URL (WebFetch), whose body/comments are read and carried into the Hunt/Judge prompts.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly fetches user-supplied non-GitHub web URLs at runtime with "WebFetch" and treats the fetched content as prior context that is injected into prompts (e.g., the example URL https://claude.com/blog/research), so remote content can directly control agent instructions.