The Agent Skills Directory

[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill instructions involve downloading and running external Node.js packages such as kordoc and korean-dart-mcp, and connecting to remote service gateways at mcp.gomdori.app and archhub-mcp.fly.dev. These components are part of the vendor's toolset for document parsing and public data retrieval.
[DYNAMIC_EXECUTION]: The setup workflow utilizes npx -y to dynamically execute MCP server packages, which fetches and runs code from the npm registry without manual confirmation. This facilitates automated configuration but bypasses user verification of the downloaded code during the initial setup.
[PRIVILEGE_ESCALATION]: The troubleshooting section includes instructions for modifying Windows Registry settings to bypass the 260-character MAX_PATH limit. While this action requires administrative privileges, it is presented as a standard solution for local file system limitations in Windows environments.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the setup of connectors that ingest data from external sources, creating a surface for potential indirect prompt injection. 1. Ingestion points: Google Drive, Notion, and local documents (HWP, PDF, DOCX) via the kordoc parser. 2. Boundary markers: None identified in the setup instructions. 3. Capability inventory: The configured connectors have network access and file system read/write capabilities. 4. Sanitization: No specific sanitization or filtering measures are described in this configuration guide.

setup-mcp-connector