dbs-theory-grounding
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's core functionality relies on processing untrusted user data to evaluate empirical judgments.
- Ingestion points: Untrusted data enters the agent context in
SKILL.md(Step 1) when extracting the "Original Proposition" and "Original Case" from the user. - Boundary markers: The skill does not employ specific delimiters (e.g., XML tags or unique string sequences) to isolate user-provided content from its internal execution instructions.
- Capability inventory: The skill acknowledges the presence of network-capable tools including
curl, web browsers, and network APIs. While it provides instructions to restrict these, they remain a target for potential injection attacks intended to force unauthorized external communication. - Sanitization: The instructions lack specific validation or escaping mechanisms for the user-supplied text before it is utilized to define search mechanisms and match theories in Step 3.
Audit Metadata