The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill's core workflow involve generating a Node.js script (build.mjs) and executing it using the drawio-ai run command. As documented in references/api.md, these scripts import the node:fs module to write the resulting diagram files to disk. Executing dynamically generated code with file system access is a high-capability operation that requires trust in the generation process.
[COMMAND_EXECUTION]: The skill frequently invokes shell commands and CLI tools. This includes the drawio-ai utility for various tasks (search, scaffold, run, validate, render) and external tools like terraform, terramate, and grep used in the workflows/from-iac.md workflow to inventory infrastructure resources.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it transforms user-supplied architectural descriptions into executable JavaScript code.
Ingestion points: User requests defining diagram components, labels, and structures (SKILL.md Step 1).
Boundary markers: The instructions do not specify any boundary markers or delimiters to separate untrusted user input from the script logic during the code generation phase.
Capability inventory: The agent uses file editing tools to create build.mjs, which is subsequently executed via drawio-ai run. The script has the capability to perform file system operations via writeFileSync.
Sanitization: There are no instructions for sanitizing or escaping user-provided strings (such as labels or icon names) before they are interpolated into the generated script, potentially allowing a malicious user to inject arbitrary JavaScript code.

drawio