The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill has a broad attack surface as it ingests and processes untrusted data from multiple sources to generate diagrams.
Ingestion points: Multiple files under scripts/adapters/ (Mermaid, CSV, Terraform, Kubernetes, SQL, OpenAPI).
Boundary markers: Data is normalized into a structured YAML specification before rendering.
Capability inventory: The skill uses Bash to run its CLI and Write to create .drawio files and sidecars.
Sanitization: It scrubbed sensitive keys (e.g., passwords, tokens) and enforces attribute allowlists in scripts/adapters/graph-projection.js.
[EXTERNAL_DOWNLOADS]: The skill is configured to download and execute external packages at runtime.
The .mcp.json file uses npx --yes to fetch and run @next-ai-drawio/mcp-server@0.4.13 from the npm registry for its optional live refinement feature.
Optional parsers like tree-sitter and es-module-lexer are loaded dynamically via import() if required by the chosen input format.
[DYNAMIC_EXECUTION]: The skill utilizes runtime execution of scripts and external processes.
The scripts/adapters/optional-python.js utility spawns Python 3.9+ processes to execute internal worker scripts (e.g., config-parser-worker.py) using isolated mode (-I) for processing Terraform and SQL data.
It uses dynamic imports for loading specific Node.js parsing libraries based on the user's input type.

drawio