katana
Pass
Audited by Gen Agent Trust Hub on Oct 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to crawl and map web targets, including inputs, forms, and script-discovered endpoints. This involves ingesting untrusted data from external websites into the agent's context.
- Ingestion points: Web targets processed during the mapping/crawling phase.
- Boundary markers: Not explicitly defined in the instructions.
- Capability inventory: Reading target content, mapping routes, and preserving crawl records.
- Sanitization: Instructions mention normalizing discovered routes and parameters, but no specific security sanitization for LLM context is described.
- [NO_CODE]: The skill contains natural language instructions but no executable scripts or command-line operations are directly provided in the SKILL.md file.
Audit Metadata