Conflict Patterns
Pass
Audited by Gen Agent Trust Hub on Jun 3, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill contains phrases often used in prompt injection attacks, such as "ignore previous instructions" and "disregard all rules." However, these are documented as behavioral patterns the skill must identify in analyzed rule sets, not as instructions for the agent to follow. This is a functional requirement for its purpose as a conflict detection tool.
- [INDIRECT_PROMPT_INJECTION]: The skill has an ingestion surface for untrusted data as it processes external rule lists.
- Ingestion points: The "Detection Process" in SKILL.md specifies that the skill receives and processes rule lists from external sources.
- Boundary markers: No specific delimiters or markers are defined for isolating the input rule text.
- Capability inventory: The skill is limited to natural language reasoning and JSON output; it lacks tools for network requests, file system access, or command execution.
- Sanitization: No explicit sanitization of the rule text is performed, but the lack of dangerous capabilities mitigates the risk.
- [SAFE]: No other security concerns such as data exfiltration, obfuscation, or persistence mechanisms were detected.
Audit Metadata