pi-agent

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

• Third-party content exposure detected (medium risk: 0.30). In the pi-web-access skill package, the runtime path /websearch → web_search(...) and /curator//search/fetch_content(...) can fetch and ingest outsider-authored text from attacker-influenced URLs, GitHub repos, and search results before the agent selects a specific item.

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

• Potentially malicious external URL detected (high risk: 0.90). The pi-web-access tool (references/pi-web-access.md) exposes fetch_content calls that at runtime fetch remote URLs (e.g., https://example.com/article, https://github.com/owner/repo, https://youtube.com/watch?v=abc) and inject that external content into the agent context, which directly controls prompts and model input.