scienceclaw-qa
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill functions as a set of instructional guidelines for the agent, defining a four-step process (Identify, Retrieve, Synthesize, Cite) to ensure scientific accuracy and rigour.
- [SAFE]: It references a taxonomy of scientific disciplines to help the agent categorize questions but does not contain executable scripts, configuration files that modify the system, or hidden commands.
- [SAFE]: A 'Zero-Hallucination Rule' is implemented, which explicitly instructs the agent to ground all factual claims in actual tool results (API calls, web search) rather than training data, serving as a safety guardrail.
- [INDIRECT_PROMPT_INJECTION]: The skill possesses a standard input surface (user questions), but its methodology requires cross-referencing multiple sources and using specialized skills for retrieval and reasoning, which mitigates the impact of malicious content embedded in user queries. Severity is considered safe given the instructional nature.
Audit Metadata