buying-role-hypotheses
Pass
Audited by Gen Agent Trust Hub on Oct 25, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill requires the agent to run a local Python script,
scripts/role_check.py, to validate formatting and check for missing content in the generated table. The script is provided within the skill package and performs benign text analysis without network access. - [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from external sources and user-pasted content.
- Ingestion points: External company product pages (fetched at runtime) and user-supplied interview or win notes (SKILL.md).
- Boundary markers: The instructions include a specific 'Golden Rule' and a directive that all fetched or uploaded text is data, not instructions, and must not be followed (SKILL.md).
- Capability inventory: The skill can perform web searches, fetch web pages, and execute a local script.
- Sanitization: The skill relies on natural language boundary instructions to prevent data from being interpreted as agent commands.
- [EXTERNAL_DOWNLOADS]: The skill may fetch information from external websites when investigating a user's company product and buyer profile, though no specific third-party domains are hardcoded.
Audit Metadata