The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill contains setup instructions that pipe remote scripts directly into a shell or interpreter (e.g., curl -fsSL https://texttocad.dev/install.sh | sh and irm https://www.texttocad.dev/install.ps1 | iex). This execution pattern is highly risky as it allows for the execution of unvetted code on the host machine.
[EXTERNAL_DOWNLOADS]: The skill's primary tool, text-to-cad, dynamically downloads binary dependencies and environments (such as MuJoCo and CoACD) during runtime. This represents an unverifiable dependency risk.
[COMMAND_EXECUTION]: The skill uses a powerful CLI tool (cad) that can execute Python scripts, perform file system operations, and manage network resources, providing a high-privilege environment that could be exploited.
[INDIRECT_PROMPT_INJECTION]: The skill's core function is to process and review complex XML robot descriptions and CAD data formats. This creates a surface where maliciously crafted files could attempt to influence the agent's logic or exploit the underlying toolchain.
Ingestion points: Robot description files (URDF, SRDF, MJCF) and CAD models (STEP, XBF) provided by the user or fetched from repositories.
Boundary markers: There are no explicit instructions for the agent to treat these file contents as data rather than instructions.
Capability inventory: The skill has access to shell execution, Python script execution, and network access for dependency retrieval.
Sanitization: While validation commands like cad urdf validate are present, they are focused on format correctness rather than security sanitization.

robot-simulation