The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/setup.py utility script executes system commands to manage the environment and install skills. The apply_plan function uses subprocess.run(command, shell=False) to execute operations approved by the user. The script implements a robust validation mechanism in _operation_error that restricts commands to a specific allowlisted prefix (npx skills add) and performs argument sanitization to prevent injection.
[REMOTE_CODE_EXECUTION]: The skill facilitates the installation of additional agent capabilities from remote repositories. The installation process uses npx skills add to pull skill definitions from sources listed in the catalog, including vendor-owned repositories and third-party sources such as mattpocock/skills.
[EXTERNAL_DOWNLOADS]: The skill documentation directs users to download and install external tools from non-standard locations. Documentation in references/google-workspace.md points to https://gogcli.sh for the gogcli tool installation.
[INDIRECT_PROMPT_INJECTION]: The installation workflow relies on the agent generating and subsequently reading an execution plan file, which presents a surface for indirect injection if not handled securely. Evidence Chain: (1) Ingestion point: plan.json file input to scripts/setup.py. (2) Boundary markers: A mandatory "write gate" (mutation-gate) requires human verification of every operation before execution. (3) Capability inventory: Subprocess execution for skill installation. (4) Sanitization: The Python script performs strict schema validation and command prefix enforcement on the plan content.

setup-rhdh-skills