The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data including URDF (XML) and Parquet files.
Ingestion points: URDF files are ingested via UrdfTree.from_file_path in SKILL.md and parsed in scripts/check_joint_mapping.py. Parquet data is ingested in scripts/check_joint_mapping.py via the pyarrow library.
Boundary markers: None identified; instructions do not specify delimiters or warnings for embedded malicious content in the data files.
Capability inventory: The skill uses Bash, Read, and WebFetch tools, providing a surface for data-driven attacks to reach system resources.
Sanitization: scripts/check_joint_mapping.py utilizes xml.etree.ElementTree for URDF parsing. This standard library component is noted to be vulnerable to XML external entity (XXE) attacks if used on untrusted input without specific safeguards.
[COMMAND_EXECUTION]: The instructions guide the user to execute shell commands using Bash, such as xacro for model conversion and uv run for executing the provided validation script. These commands are consistent with the skill's stated purpose of processing robotics data.
[EXTERNAL_DOWNLOADS]: The skill facilitates downloading robot descriptions and mesh assets from well-known technology organizations and official service repositories, such as Google DeepMind, Universal Robots, and the author's own infrastructure.

rerun-urdf