research
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [SAFE]: The skill establishes a structured framework for research and discovery, emphasizing evidence over authority without introducing executable code or dangerous dependencies.
- [INDIRECT_PROMPT_INJECTION]: The skill provides explicit instructions to treat all external inputs as untrusted evidence. It mandates that instructions embedded within webpages or documents must not be followed as authority. Evidence chain:
- Ingestion points: External sources including webpages, documents, issues, repository text, and tool outputs (SKILL.md, references/discovery-protocol.md).
- Boundary markers: Present. Instructions define an 'Untrusted-source boundary' and 'Untrusted-source hard boundary' to distinguish evidence from authority.
- Capability inventory: Implicit retrieval and processing of external data via platform tools.
- Sanitization: Behavioral requirement to record redirection attempts as source content instead of executing them.
- [DATA_EXFILTRATION]: The protocol includes specific rules to avoid copying restricted or confidential material into public artifacts and to respect access limitations and provenance of source data.
Audit Metadata