The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides configuration instructions to download and run the exa-mcp-server package from the NPM registry using npx.
Evidence: "command": "npx", "args": ["-y", "exa-mcp-server"] in SKILL.md.
Context: The package is the official MCP server for Exa (exa.ai), a well-known technology service.
[COMMAND_EXECUTION]: The setup instructions include a shell command configuration for the agent platform to execute a third-party package.
Evidence: The MCP configuration block in SKILL.md defines a command and arguments for runtime execution.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted content from the external web and code repositories, creating an attack surface where malicious instructions embedded in search results could influence agent behavior.
Ingestion points: The web_search_exa and get_code_context_exa tools in SKILL.md return data from arbitrary third-party websites and code hosts.
Boundary markers: Not present; the skill does not specify delimiters or instructions to ignore embedded prompts in the retrieved data.
Capability inventory: The skill itself uses search tools; however, the agent environment it operates in may have additional capabilities like file writing or command execution.
Sanitization: Not present; retrieved content is processed directly by the model without escaping or validation filters described in the instructions.

exa-search