The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands for system preflight, software installation, and verification using tools like uv, pip, and docker. Commands are dynamically sourced from the project's documentation files.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests text from local files (e.g., docs/source/setup/installation/index.rst) and interprets them as commands to be executed. This pattern is vulnerable if the documentation files are modified to include malicious instructions.\n
Ingestion points: Documentation files like index.rst, manage_asset_downloads.rst, and docker/index.rst.\n
Boundary markers: Commands are extracted from documentation sections and executed without isolation or verification against a whitelist.\n
Capability inventory: Full shell access, including sudo capabilities and the ability to write to the user's home directory (~/.isaaclab/).\n
Sanitization: There is no documented process for validating the integrity or safety of the commands extracted from documentation files before execution.\n- [PRIVILEGE_ESCALATION]: The installation process incorporates sudo steps for system configuration. Although a confirmation is required before execution, this allows for the execution of high-privilege operations derived from documentation content.

isaaclab-installing-isaac-lab