The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches skill configuration, metadata, and the live skill index from the vendor's official GitHub repositories.
Evidence: Uses curl and gh api to read skills-config.json and skill SKILL.md files from raw.githubusercontent.com/open-edge-platform/skills and api.github.com/repos/open-edge-platform/skills in references/DISCOVERY.md.
[REMOTE_CODE_EXECUTION]: Dynamically installs and executes delegate skills from the open-edge-platform organization using the npx skills CLI.
Evidence: The core functionality in SKILL.md Step 5 uses npx skills add open-edge-platform/skills --skill <skill-name> to pull remote skill definitions into the environment.
[COMMAND_EXECUTION]: Executes shell commands to manage the environment and invoke the delegated building blocks.
Evidence: Utilizes bash tools like npx, git, gh, ls, and curl to discover, install, and run skills as defined in SKILL.md and references/DISCOVERY.md.
[PROMPT_INJECTION]: The skill processes untrusted user input (business objectives) which influences the selection and configuration of tools and remote skills, representing an indirect prompt injection surface.
Ingestion points: User-provided business objectives in SKILL.md Step 1.
Boundary markers: The skill implements a mandatory human-in-the-loop checkpoint in Step 4 ("WAIT for confirmation") before any installation or build action is taken.
Capability inventory: The agent has access to bash, git, gh, and npx, enabling file system access and network operations.
Sanitization: No explicit content sanitization is described in the orchestrator instructions, relying on the user's manual review of the proposed plan.

metro-ai-apps-builder