icp-builder
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external LinkedIn profiles and social posts which is then written to local configuration files used by other GTM skills.\n
- Ingestion points: SKILL.md uses person_enrich and social_post_list_live tools to fetch profile and post data from external sources.\n
- Boundary markers: The skill includes a manual verification step where the inferred persona is shown to the user for manual confirmation and correction before any files are written to disk.\n
- Capability inventory: The skill utilizes callTool for network data retrieval and has local file-write access to create config/persona-profile.md and config/gtm-config.md.\n
- Sanitization: No explicit technical sanitization or escaping of the ingested post content is defined in the scripts, relying instead on instructions to the agent to filter specific phrases (no-slop rule).\n- [EXTERNAL_DOWNLOADS]: The skill documentation refers to the Crustdata MCP server and official releases hosted on GitHub.\n
- Referenced source: https://install.crustdata.com/mcp\n
- Referenced source: https://github.com/crustdata/skills/releases\n
- These references target the official resources of the skill author, crustdata.
Audit Metadata