The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches runtime API documentation and usage guides from an external URL when local documentation is unavailable.
Evidence: Fetches content from https://durable-agents-docs.netlify.app/ (a well-known service provider, Netlify).
Context: This is used to provide the LLM with authoritative reference material for the @electric-ax/agents-runtime during the design phase.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a code generator that ingests untrusted user descriptions to produce TypeScript entity files.
Ingestion points: User input provided in the 'Elicit' phase (Phase 1) describing the desired agent behavior.
Boundary markers: While explicit delimiters are not used in the elicitation phase, the workflow enforces a strict separation between description, design, and implementation.
Capability inventory: The skill can write exactly one file to the filesystem (entities/<type-name>.ts) and read from documentation directories or the specified Netlify URL.
Sanitization: The skill contains a mandatory 'Review' phase (Phase 4) that audits the proposed design against a comprehensive security checklist (e.g., checking for credential leakage, ensuring deterministic IDs, and verifying tool configurations) before any code is implemented. This significantly mitigates the risk of the agent being tricked into generating malicious or insecure code.

designing-entities