The Agent Skills Directory

[SAFE]: No malicious patterns or security violations were detected. The skill's operations align with its stated purpose of document management and follow best practices for handling sensitive transfer tokens.\n- [PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it processes external document content and comments from the Nativ Docs MCP server. These inputs could potentially contain instructions aimed at the agent. The skill mitigates this by using sandboxed environments for HTML and sanitization for Markdown.\n
Ingestion points: get_document (delivery: inline) and list_comment_threads mentioned in SKILL.md.\n
Boundary markers: Not explicitly defined for prose analysis, but instruction 6 advises the agent against unnecessary processing of file content.\n
Capability inventory: Generation of curl commands for local execution, network writes to mcp.nativdocs.co via create_document and update_document.\n
Sanitization: Markdown sanitization and HTML sandboxing (iframe with no network access) are explicitly specified in the document capabilities section.\n- [COMMAND_EXECUTION]: The skill facilitates document synchronization by generating curl commands for the user to execute. This allows for direct transfer of files up to 10MB between the local disk and the Nativ server, which is a core functional design for efficient data movement.\n- [EXTERNAL_DOWNLOADS]: The skill relies on an external service at https://mcp.nativdocs.co/mcp for document storage and retrieval. This is a legitimate functional requirement for the Nativ Docs integration.

nativ-docs