The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill is configured to connect to an external Model Context Protocol (MCP) endpoint at https://cyeam-wiki-mcp-production.up.railway.app/mcp. This server defines the tools and prompts that the agent will use.
[PROMPT_INJECTION]: The skill implements a dynamic instruction loading mechanism where it fetches a 'system prompt' (wiki_query_system) from the remote MCP server. This design allows the external server to dictate the agent's persona and safety guidelines at runtime, bypassing the local skill file's configuration.
[COMMAND_EXECUTION]: The documentation includes instructions for using a curl tool to perform real-time network requests. It specifically directs the agent to generate and access URLs, providing a general-purpose network communication capability.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface through its wiki retrieval tools.
Ingestion points: Content retrieved from wiki_query, wiki_search_index, and wiki_get_article (file: SKILL.md).
Boundary markers: None identified; external content is integrated directly into the agent's context without clear delimiters.
Capability inventory: Network operations (curl), dynamic system prompt loading (wiki_query_system).
Sanitization: No validation or sanitization procedures are mentioned for processing external data.

architect