The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands to perform project initialization, dependency management, and quality checks.
Commands are executed via Python's subprocess module in scripts like preflight.py, blueprint.py, and doctor.py to gather project state and run tools.
All mutating or potentially risky commands are categorized and require explicit user approval through a generated blueprint before they can be executed by the agent.
The system includes an internal catalog (data/command-groups.json) used to identify and block destructive commands (e.g., rm -rf, git clean) and commands that might access sensitive files.
[EXTERNAL_DOWNLOADS]: Project setup capabilities involve downloading and running scaffolding tools from remote registries.
The skill supports fetching tools like next-app, astro, shadcn, and nx via package managers such as pnpm, npm, and uv.
These downloads are standard for the development ecosystems supported by the skill and are disclosed to the user within the project blueprint for explicit approval.
[REMOTE_CODE_EXECUTION]: The skill uses npx to fetch and execute CLI tools such as @fission-ai/openspec during project governance setup.
This execution pattern is part of the 'Apply' phase, which is restricted until the user has reviewed and approved the blueprint containing the specific commands.

new-project