The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill relies on an external package, drawio-ai-kit, hosted on the author's GitHub repository. The instructions include a specific safeguard telling the agent not to install the package automatically (Never run npm i -g yourself), ensuring the user retains control over their global environment.
[COMMAND_EXECUTION]: The skill makes extensive use of the drawio-ai CLI tool to perform searches, scaffolding, validation, and rendering. These commands are localized to the diagram generation task.
[INDIRECT_PROMPT_INJECTION]: The skill involves a subagent pattern where user-provided architecture requests are passed into a prompt to generate code.
Ingestion points: User architectural requests are interpolated into the subagent prompt in SKILL.md.
Boundary markers: The skill uses clear instructions for the subagent to return a specific block of metadata (DRAWIO:, PNG:, VALIDATE:) and forbids invoking other skills.
Capability inventory: The agent can run shell commands, write .mjs files, and read rendered images.
Sanitization: The skill relies on the drawio-ai validate command to verify the integrity and correctness of the generated diagram logic before final delivery.
[DYNAMIC_EXECUTION]: The workflow involves generating a temporary script (build.mjs) based on a template, which is then executed using Node.js to render the diagram. This is a standard operational pattern for this skill's primary purpose of programmatic diagram generation.

drawio-azure