The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The workflow templates download development tools including uv, python-build-standalone, bun, and tailwindcss from their official GitHub release pages.
[REMOTE_CODE_EXECUTION]: The skill provides instructions for installing its CLI tool using a remote script and includes Dockerfile templates that pipe remote archives directly into extraction tools during the build process.
[DYNAMIC_EXECUTION]: The agent runtime uses dynamic imports to load tool definitions from a local directory, which is a standard architecture for modular agent capabilities.
[COMMAND_EXECUTION]: The skill is designed to manage and execute custom code in Shell, Python, and TypeScript, which is the core functionality of the workflow platform.
[INDIRECT_PROMPT_INJECTION]: The agent template possesses an attack surface for indirect injections as part of its design for processing external triggers.
Ingestion points: The agent's entry point reads untrusted trigger data or HTTP request bodies from standard input.
Boundary markers: Authentication headers are verified and filtered by the platform before script execution to prevent header spoofing.
Capability inventory: Workflow steps are capable of executing shell and script code, performing network operations, and writing to persistent storage volumes.
Sanitization: The skill implements sanitization for conversation identifiers and relies on platform-level header stripping for security.

building-workflows