From discovery to execution
Recipes your agent can use
Each recipe is both a human-readable guide and a machine-readable plan with ordered capabilities, tool choices, and fallbacks.
How to use a recipe
A recipe is a starting architecture, not a copy-paste guarantee. Review every input, output, fallback, credential boundary, and provider source. The JSON endpoint mirrors the human-readable plan so an agent or build system can load the same ordered steps.
Before production use, replace example assumptions with the constraints of your own workload. Pin the provider and API version, define timeouts and budgets, test each fallback, scope credentials per step, and store enough trace data to explain the final result. Re-evaluate the tool choices when requirements or provider behavior changes.
4agent uses the terminology and machine-readable interface conventions documented by the Model Context Protocol specification and the OpenAPI Specification.