business-model
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill promotes secure behavior by requiring the agent to respect authorization boundaries, verify outputs through tests rather than just file creation, and seek user approval before using external tools that incur costs. It specifically mandates data minimization when interacting with external services.
- [INDIRECT_PROMPT_INJECTION]: The skill utilizes tools for external data collection (sandbase_run), creating a potential surface for indirect prompt injection through external evidence gathering and web scraping.
- Ingestion points: External evidence, search, and scraping results obtained via sandbase_run as described in SKILL.md.
- Boundary markers: Workflow step 2 requires restating outcomes and constraints, and step 4 requires verification against source material, though no specific prompt delimiters are defined.
- Capability inventory: Network-enabled data discovery, inspection, and execution tools (sandbase_discover, sandbase_run).
- Sanitization: The skill relies on source comparison, human-in-the-loop confirmation for costs, and manual validation rather than automated content filtering.
Audit Metadata