The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data which could contain adversarial instructions.
Ingestion points: Data is ingested via web_search_live (search engine results), web_enrich_live (external website content), and social_post_search_live (social media posts) as described in SKILL.md recipes.
Boundary markers: The instructions do not specify explicit delimiters or 'ignore' instructions for the content fetched from the web or social media.
Capability inventory: The skill has the capability to execute JavaScript via the execute tool, perform further network searches, and access contact enrichment tools.
Sanitization: The skill implements basic junk filtering by excluding profiles with low connection counts or specific titles (e.g., 'advisor', 'investor') as outlined in the champion tracker and person_search recipes.
[DYNAMIC_EXECUTION]: The skill instructs the agent to generate and run JavaScript code to interact with tools.
Evidence: The SKILL.md file defines that 'All Crustdata data tools run inside execute({ code }) — a plain-JavaScript script'. It provides specific JS templates for the agent to follow when performing company and person searches.
[EXTERNAL_DOWNLOADS]: The skill fetches resources from remote vendor-owned URLs.
Evidence: The skill fetches company logos and person profile pictures from basic_info.logo_permalink and basic_profile.profile_picture_permalink. It also references the vendor's MCP server at https://install.crustdata.com/mcp for tool execution.

sales-prospecting