The Agent Skills Directory

[SAFE]: The skill implements a deterministic planning workflow where natural language mechanical requirements are converted into structured JSON data. All processing is performed locally using Python scripts (scripts/plan_from_brief.py and scripts/sw_execution_outline.py) that utilize regular expressions for parameter extraction, avoiding any dynamic execution of user input.
[SAFE]: The system prompts (prompts/solidworks_parametric_planner.md) establish strict boundaries by explicitly prohibiting the AI from generating VBA or Python macros. It forces the output into a structured JSON format governed by a schema, which reduces the risk of the agent performing unauthorized actions in a CAD environment.
[SAFE]: No external network communication, hardcoded credentials, or remote code execution patterns were detected. The skill uses standard Python libraries for file operations and refers to internal modular subskills for actual SolidWorks integration, which were not part of this analysis but the planning logic itself is self-contained and secure.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface as it processes natural language text from users.
Ingestion points: The scripts/plan_from_brief.py script reads untrusted user input via command-line arguments or text files.
Boundary markers: The skill uses a detailed system prompt that provides instructions to ignore embedded code and strictly adhere to a JSON schema.
Capability inventory: The scripts are limited to file I/O operations and do not have access to network or shell execution capabilities.
Sanitization: The parsing logic utilizes specific regex patterns to filter for dimensions and mechanical specifications, ignoring irrelevant or potentially malicious text segments.

solidworks-vibecad