The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The search script scans the user's home directory (~/.mcp-auth/) to locate and read authentication tokens stored in JSON files. It also provides instructions for manually adding API keys to local configuration files.
[REMOTE_CODE_EXECUTION]: The script uses npx to dynamically download and execute the mcp-remote package from the npm registry if valid authentication tokens are not found locally.
[COMMAND_EXECUTION]: The skill executes several shell commands to perform its core functions, including curl for API requests, jq for JSON manipulation, and find for scanning credential directories.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted search queries and retrieves content from external web pages, which is a common vector for indirect prompt injection.
Ingestion points: The script takes user-supplied search queries via the JSON_INPUT argument in scripts/search.sh.
Boundary markers: No explicit markers or instructions are used to delimit or neutralize embedded instructions in the search results.
Capability inventory: The skill has the capability to perform network requests (curl) and process structured data (jq).
Sanitization: The script does not perform sanitization or filtering on the content returned from the Tavily search results before passing it to the agent.

tavily-search