The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses npx to execute the MCP Inspector (mcpdo) and Chrome DevTools MCP servers. It also employs curl for manual JSON-RPC messaging over HTTP.
[EXTERNAL_DOWNLOADS]: Fetches and runs the official @modelcontextprotocol/inspector (pinned to version 2.10.1) and chrome-devtools-mcp (latest version) packages from the npm registry. These are official tools from trusted organizations.
[REMOTE_CODE_EXECUTION]: Downloads and executes packages from the npm registry at runtime via npx. This involves the official MCP Inspector client and the Google Chrome DevTools MCP server.
[INDIRECT_PROMPT_INJECTION]: The skill processes data returned from external MCP servers, creating a potential attack surface for indirect prompt injection.
Ingestion points: Tool call results and JSON-RPC responses retrieved via mcpdo and curl, as documented in references/inspector.md and references/protocol.md.
Boundary markers: There are no specific delimiters or instructions to ignore embedded commands within server output.
Capability inventory: Capabilities include shell command execution via mcpdo and npx, and network operations via curl.
Sanitization: The skill does not provide instructions for sanitizing or escaping external server data before processing.
[DATA_EXFILTRATION]: The instructions involve reading local configuration files (e.g., claude_desktop_config.json, .mcp.json, and ~/.codex/config.toml) to extract server commands, arguments, and environment variables. This data access is necessary for the skill's primary function of managing MCP connections.

mcp-access