The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The script scripts/extract.sh uses npx -y mcp-remote, which dynamically downloads a package from the npm registry during execution.
[REMOTE_CODE_EXECUTION]: Executing mcp-remote via npx involves running code fetched from a remote repository at runtime to handle the OAuth flow.
[COMMAND_EXECUTION]: The script performs subprocess execution of curl, jq, npx, and find to perform operations, file searches, and network requests.
[DATA_EXFILTRATION]: The script accesses the user's home directory at ~/.mcp-auth/ to locate sensitive authentication tokens and transmits discovered tokens to the external endpoint https://mcp.tavily.com/mcp.
[PROMPT_INJECTION]: The skill processes content from arbitrary external URLs, creating a surface for indirect prompt injection attacks.
Ingestion points: The urls array in the JSON input passed to scripts/extract.sh.
Boundary markers: No specific delimiters or "ignore instructions" warnings are applied to the extracted web content before it is returned to the agent.
Capability inventory: The script has the capability to read local files (~/.mcp-auth/) and make network requests (curl, npx).
Sanitization: No sanitization, escaping, or filtering is performed on the content retrieved from external web pages.

tavily-extract