The Agent Skills Directory

[COMMAND_EXECUTION]: The Python CLI in tools/src/arch_builder/cli.py uses subprocess.run to orchestrate internal helpers, including a Node.js rendering script and an AWS icon build script. The skill also provides configuration for an optional MCP server initialized via npx in vendor/drawio/.mcp.json.
[EXTERNAL_DOWNLOADS]: The skill references external packages from established registries. It optionally fetches the @next-ai-drawio/mcp-server package via NPM and lists requirements like python-hcl2 and sqlglot for parsing infrastructure files.
[DYNAMIC_EXECUTION]: JavaScript logic in the vendor/drawio directory uses dynamic import() calls to load specific parsing engines like tree-sitter and es-module-lexer only when needed for specific input formats.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from Terraform, Kubernetes, and SQL sources, creating a potential injection surface. 1. Ingestion points: Data is read by specialized adapters in kubernetes.js, terraform-config.js, and sql-ddl.js. 2. Boundary markers: Input is processed through formal parsers (YAML, JSON, HCL, SQL) before being used in the diagramming pipeline. 3. Capability inventory: The skill can execute local commands and perform file writes for diagram generation. 4. Sanitization: It features extensive filtering logic in artifacts.js to strip sensitive information such as secrets and credentials from processed metadata.
[OBFUSCATION]: The skill includes logic to encode and decode standard draw.io Base64-compressed XML and image data URIs, which are required for interoperability with the draw.io file format.

arch-builder