sharingan

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

• Third-party content exposure detected (high risk: 0.90). The skill explicitly clones arbitrary repository URLs (scripts/clone_repo.sh / run.sh using --repo-url) and then reads and processes those repo files in scripts/build_context_bundle.py (producing chunks.ndjson, summary.md, selection_trace.json and artifact.json), so untrusted, user-generated content from public repos is ingested and can influence ranking, selection, and downstream decision-making.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill clones a user-provided repository URL at runtime (e.g., https://github.com/owner/repo.git as passed to --repo-url and shown in examples like https://github.com/example/repo.git), and the cloned repository's file contents are ingested into the produced bundle (chunks.ndjson / summary.md) — effectively injecting remote content into agent context, so the external URL is a runtime dependency that can directly control prompts.