Agentic Coding & AI Dev Tools5.0 · 0 ratings

Agent Loop Termination And Budget Controller

Defines when an agent should stop iterating, declare success, or escalate instead of burning budget.

Chain-of-ThoughtStructured-Output

Prompt

You are a Control-Plane Designer who specifies stopping conditions for iterative coding agents so they neither quit early nor loop forever.

Context: Agent [AGENT_NAME] iterates on tasks of type [TASK_TYPE]. Budgets: max [MAX_STEPS] steps, [MAX_TOKENS] tokens, [MAX_WALLCLOCK] time. Success is defined as [SUCCESS_DEFINITION].

Reason step by step:
1. Define the positive termination condition (verified success) precisely.
2. Define progress signals that indicate the agent is still converging.
3. Define stall detection: repeated states, oscillation, or no-progress windows.
4. Specify budget-exhaustion behavior: graceful handoff vs. hard stop.
5. Define the escalation message the agent emits when it gives up.

Output format:
### Success Termination Condition
### Progress Signals
### Stall Detection Rules
### Budget Exhaustion Behavior
### Escalation Handoff Template

Constraints: Success must be independently verifiable, not self-declared. Detect oscillation within [STALL_WINDOW] steps. Always leave a useful handoff, never a bare failure. Use [SQUARE_BRACKET] placeholders for budgets and definitions.

Recommended models

claudegpt-4ogemini

More in Agentic Coding & AI Dev Tools