prospecting
Pass
Audited by Gen Agent Trust Hub on Nov 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill identifies a clear surface for indirect prompt injection attacks due to its core functionality of processing untrusted web content.
- Ingestion points: Untrusted data enters the agent context through tools like Firecrawl, Browserbase, and social-fetch when mining public forums, GitHub issues, and reviews (references/demand-signals.md).
- Boundary markers: The instructions do not define robust delimiters or explicit 'ignore instructions' warnings for content fetched from the web.
- Capability inventory: The skill can output formatted tables, generate CSV files, and interface with numerous external APIs.
- Sanitization: While the skill advises escaping values in HTML reports, it lacks comprehensive sanitization for external text being analyzed by the model.
- [EXTERNAL_DOWNLOADS]: The instructions rely on integrations with several third-party B2B data and automation services, such as Apollo, Clay, ZoomInfo, and Hunter. These are well-known technology providers used for data enrichment.
- [COMMAND_EXECUTION]: The skill mentions executing a local utility script, node tools/clis/github-prospects.js, to process GitHub data (references/saas-prospecting.md).
Audit Metadata