The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill relies on an external Python package, cadgen (version 0.7.20), which is downloaded and executed at runtime from a public registry using the uvx tool.
Evidence: SKILL.md defines the cadgen and python commands to use uvx --no-config --managed-python --python 3.13 --from cadgen==0.7.20.
[COMMAND_EXECUTION]: The agent is instructed to execute shell commands for robot description validation, generating snapshots, and launching a local CAD viewer server.
Evidence: The skill utilizes commands such as cadgen srdf validate, cadgen snapshot, and cadgen viewer --host 127.0.0.1 --json --detach throughout the instruction set.
[INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface where it consumes data from external URDF files to extract kinematic information, which is then used to generate complex SRDF semantics.
Ingestion points: As described in SKILL.md and references/srdf-workflow.md, the agent must 'Extract the URDF table' from paired .urdf files to identify links and joints.
Boundary markers: While the workflow involves a structured 'planning ledger' (documented in references/planning-ledger.md), the skill lacks explicit prompt delimiters or sanitization warnings for the content ingested from external description files.
Capability inventory: The skill is capable of executing shell commands through the cadgen tool and writing .srdf XML files to the local file system.
Sanitization: The workflow includes a mandatory cross-validation step using cadgen srdf validate to ensure the generated SRDF is consistent with the URDF structure before concluding the task.

srdf