The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads and installs the cad-ai-agent utility from the author's GitHub repository (github.com/jeremylongshore/cad-ai-agent.git). The installation process uses a specific git commit hash (6393e61869187eec7416f7fe54bd2cec861ad39a), which is a security best practice for ensuring the integrity of the downloaded code.
[COMMAND_EXECUTION]: The skill relies on executing the cad-analyze and cad-revision shell commands to process drawings. These commands are executed locally via the Bash tool to generate analysis reports in JSON format.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it processes untrusted DXF files provided by users.
Ingestion points: DXF files are ingested through the Read and Glob tools.
Boundary markers: The skill specifies that output from the CLI tools should be parsed as structured JSON before being reported in prose, which acts as a data-handling boundary.
Capability inventory: The skill utilizes shell execution capabilities (Bash) to run the drawing analyzer.
Sanitization: The risk is mitigated by the use of a deterministic, non-LLM CLI tool for the primary data extraction, preventing untrusted content in the CAD file from directly influencing the agent's instruction stream.

cad-dxf-agent