The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches the official Docker installation script from Docker's public domain for environment setup.
Evidence: curl -fsSL https://get.docker.com | sh in templates/scripts/deploy.sh.
[COMMAND_EXECUTION]: The generated infrastructure utilizes shell commands for container orchestration and automated testing of agent skills.
Evidence: Use of subprocess.run to call the claude CLI in templates/demo-container/skill-test.py.
Evidence: Use of the --dangerously-skip-permissions flag in templates/demo-container/entrypoint.sh and autoplay.sh to facilitate automated demo interactions in a sandboxed environment.
[DATA_EXPOSURE]: Implements secure credential management practices for generated demo environments.
Evidence: Sensitive API tokens and OAuth tokens are passed to demo containers via temporary environment files with restricted file system permissions (0o600).
Evidence: Uses industry-standard .env templates and gitignore rules for credential isolation.
[PROMPT_INJECTION]: The generated demo infrastructure includes an indirect prompt injection surface where scenario files are processed as agent input.
Ingestion points: Prompts are loaded from user-defined .prompts files in the scenarios/ directory.
Boundary markers: None present in the generated templates.
Capability inventory: The container environment has permissions to execute CLI commands and make network requests to configured product APIs.
Sanitization: No explicit sanitization or filtering is performed on the prompt content before processing by the agent.

demo-scaffolding