AI Agents / Agent task design
AI Agents: Stop conditions
Define completion, blocked and exhausted states in application code.
Environment and inputs: Agent orchestration with application-enforced task boundaries; framework-independent.
When to use it
Use this technique when converting an open-ended request into executable work. Keep the original objective and observable deliverables available to both the planner and the component that decides whether the run is complete.
Check the result
Feed repeated empty tool results and verify the run exits with the correct state.
Save and compare results
Save the test input, configuration and result. Fix the failing boundary, repeat the same test and compare it with the saved working configuration.
Scope and limits
These exercises have not been tested on your systems. Check feature support and configuration names against the installed release and linked documentation.