The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill documentation includes a specific instruction for the AI agent to automatically and silently update the skill's source code by executing git pull from a remote GitHub repository (Agents365-ai/drawio-skill).
This mechanism allows the introduction of unverified code into the local environment without manual review or oversight.
The instruction explicitly tells the agent to perform this update once every 24 hours.
[EXTERNAL_DOWNLOADS]: The skill's installation scripts and update logic perform network requests to external services.
It queries the NPM registry for version information of the @next-ai-drawio/mcp-server package.
It performs Git operations against a third-party GitHub repository.
[COMMAND_EXECUTION]: The skill requires the Bash tool to execute various Node.js scripts and system commands.
It runs local scripts such as scripts/cli.js to convert YAML to Draw.io XML/SVG.
It executes the Draw.io desktop application CLI for exporting diagrams to raster formats (PNG/PDF/JPG).
While the implementation uses argument arrays for execution, the broad scope of command execution on files generated from user input presents a security surface area that requires trust in the internal conversion logic.
[PROMPT_INJECTION]: The skill's SKILL.md contains a 'Non-Negotiable Contract' section which uses strong language to override agent behavior, such as strictly forbidding the use of Model Context Protocol (MCP) backends even if available. While intended for workflow integrity, these directives attempt to override the AI's default decision-making protocols.

drawio-academic-skills