The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources, which creates a potential surface for indirect prompt injection.
Ingestion points: Data is ingested from external web sources via web_search_live (snippets and AI summaries) and web_enrich_live (content from GitHub commit patches, such as https://github.com/OWNER/REPO/commit/SHA.patch).
Boundary markers: The skill does not explicitly define delimiters or specific 'ignore' instructions to separate retrieved data from agent instructions, increasing the risk of the agent misinterpreting data as commands.
Capability inventory: The agent has extensive capabilities including executing JavaScript via the execute tool, performing network requests via callTool, and writing progress state to the local file system.
Sanitization: No specific sanitization or filtering logic is described for the raw content fetched from external URLs before processing.
[DYNAMIC_EXECUTION]: The skill utilizes runtime generation and execution of JavaScript code to coordinate tool calls.
All Crustdata tool interactions are encapsulated within execute({ code }) scripts. These scripts are generated based on templates to handle batching, parallel processing of requests, and complex verification logic.
While this is the established method for using the Crustdata MCP server, it involves the agent assembling and running executable code dynamically.
[EXTERNAL_DOWNLOADS]: The skill fetches information from external repositories and search engines to resolve contact details.
It specifically targets GitHub to retrieve raw patch files for email extraction from technical profiles.
These external references target well-known services (GitHub, professional networks) and are documented as part of the contact verification and enrichment logic.

email-enrichment