search-leads

Pass

Audited by Gen Agent Trust Hub on Jul 27, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill focuses on lead generation and business research using standard MCP tools (Anysite, Firecrawl) and built-in web search. All operations align with the stated purpose.
  • [DATA_EXPOSURE]: The skill searches for local Ideal Customer Profile (ICP) configuration files using glob patterns (**/ICP_MODEL.md, **/ICP.md). This access is restricted to specific filenames and is used to initialize search parameters provided by the user.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources, including LinkedIn profiles and company websites scraped via Firecrawl.
  • Ingestion points: SKILL.md (Step 2: Research Context; Step 4: Enrich Top Candidates) reads data from LinkedIn and external URLs.
  • Boundary markers: None identified; external content is processed directly for scoring and outreach generation.
  • Capability inventory: The skill uses network search and scraping tools. It does not possess file-write, subprocess execution, or shell access capabilities.
  • Sanitization: No explicit sanitization or instruction-ignoring delimiters are used when processing scraped profile data.
  • Risk: While vulnerable to instructions embedded in web content (e.g., a malicious LinkedIn bio), the lack of high-privilege tools limits the potential impact to prompt subversion within the current session.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 27, 2026, 07:51 PM
Security Audit — agent-trust-hub — search-leads