The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requests Bash(*) access in SKILL.md, granting the agent the ability to execute arbitrary shell commands. While aligned with the functional requirements of ROS development (e.g., using ros2, colcon, and rosdep), this represents a significant permission level.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data such as package names, node types, and dependency lists (e.g., in SKILL.md and README.md process examples) and interpolates them into generated code and configuration files. There are no explicit sanitization steps or boundary markers to prevent malicious instructions within these inputs from affecting the generated artifacts.
Ingestion points: User-supplied parameters packageName, nodeName, and dependencies in the create-package and generate-node operations.
Boundary markers: Absent in the code generation templates.
Capability inventory: Bash(*), Write, and Edit tools are available to the skill across all files.
Sanitization: No evidence of input validation or escaping for shell commands or Python code generation.
[DYNAMIC_EXECUTION]: The skill is designed to generate Python source code for ROS nodes and launch files (e.g., generate-launch, create-node). The creation and subsequent execution of these generated files at runtime constitutes dynamic execution of content derived from user-controlled inputs.
[EXTERNAL_DOWNLOADS]: The documentation in README.md recommends the installation of an external package robotmcp via pip. While this is presented as a functional requirement for MCP integration, users should verify the package before installation.

ros-integration