The Agent Skills Directory

[COMMAND_EXECUTION]: The skill guides the user through running shell commands to scaffold new projects using npm create, manage development lifecycles via npm start and npm run verify, and install IDE-specific plugins as documented in references/generator.md.
[EXTERNAL_DOWNLOADS]: The project generator downloads the codemodekit package from the npm registry. Examples in references/server-api.md also include downloading and running Docker containers from well-known registries such as the GitHub Container Registry.
[REMOTE_CODE_EXECUTION]: Documentation in references/generator.md provides examples for wrapping and executing remote tools using uvx, which downloads and runs packages from external repositories.
[COMMAND_EXECUTION]: The skill implements a run_typescript capability that executes code within a QuickJS sandbox. Additionally, the verification workflow in references/verification.md allows executing local code files defined in environment variables.
[PROMPT_INJECTION]: The skill processes untrusted project data and integrates with external APIs, presenting an indirect injection surface.
Ingestion points: The skill ingests repository files, package configurations, and outputs from integrated MCP servers in SKILL.md and references/verification.md.
Boundary markers: The architecture utilizes a QuickJS sandbox to isolate model-authored code and recommends explicit tool authorization policies in references/policy-and-security.md.
Capability inventory: The environment supports shell command execution, network operations, and filesystem interactions across various scripts.
Sanitization: The skill promotes the use of JSON Schema for input validation and provides a safe ToolError mechanism to prevent leaking sensitive host information.

build-codemodekit-server