The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches its skill index (skills-config.json) and specific skill definitions (SKILL.md) from the vendor's official GitHub repositories (open-edge-platform/skills). This is a standard mechanism for keeping the agent's capability catalog synchronized with the platform's latest offerings.
[COMMAND_EXECUTION]: Utilizes system tools including bash, git, and gh to manage local files, query the GitHub API, and interact with the local development environment.
[REMOTE_CODE_EXECUTION]: Employs npx skills add to dynamically download and install new skills from the vendor's repositories. This capability is managed through a multi-step procedure where the agent must present a plan and wait for the user to provide explicit approval before any code is added or executed.
[PROMPT_INJECTION]: The skill processes untrusted business objectives from users and ingests external metadata from remote skill repositories, which constitutes a surface for indirect prompt injection.
Ingestion points: User-provided business objectives and remote SKILL.md files fetched via curl from GitHub.
Boundary markers: The skill is instructed to separate technical execution details from the business-level conversation and to present inferred technology as "decisions made" rather than suggestions.
Capability inventory: Access to bash, git, gh, and the ability to install new Node.js-based skills via npx.
Sanitization: The skill relies on a mandatory human-review step (Step 4) where the user must explicitly approve the build plan and inferred technical parameters before any capabilities are exercised.

metro-ai-apps-builder