ingest

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly opens a user-approved external domain (e.g., arxiv.org) at runtime and downloads/normalizes that content into the vault which is then injected into the model context, so external content can directly influence prompts (flagged URL: arxiv.org).

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

• Attempt to modify system services in skill instructions detected (low risk: 0.30). The skill instructs runtime changes to the machine's network isolation (open/seal domains), which modifies system state, but it requires explicit user consent, emphasizes not opening network on its own, and does not request sudo, edit system files, or create users, so the risk is limited.