Skip to content
Awaken

Configure Stop Policies

Stop policies keep an agent from running longer than intended. Configure them from the agent editor first; use API-level policy objects only when you are automating the same change.

What you click

Section titled “What you click”
Agent editor Basics tab with max rounds, max continuation retries, reasoning effort, system prompt, save controls, and preview chat.
Use Basics for max rounds and retries first; open Advanced only when you need raw policy review.
  1. Open Agents and choose the agent.
  2. In Basics, set the common bound first: max rounds.
  3. If the agent needs stricter conditions, open Advanced and review the stop policy shape exposed by the server.
  4. Click Validate.
  5. Preview two scenarios: one that should complete normally, and one that should stop.
  6. Save only when both scenarios behave correctly.
  7. Run an eval fixture for loops or time-sensitive behavior.

Policy choices

Section titled “Policy choices”
RiskUI choiceHow to verify
Repeating the same planLower max rounds or add loop detectionPreview a prompt that used to loop
Long-running tool chainAdd timeout or tool-stop conditionPreview with a tool-heavy task
Token runawayAdd token budgetEval fixture with long context/tool output
Error stormAdd consecutive-error boundPreview with a failing tool or provider
Stop after a known terminal toolAdd stop-on-tool conditionPreview a task that calls that tool once

Operate after saving

Section titled “Operate after saving”
Agents list with runtime inference statistics.
Watch run and latency signals after tightening limits.
Agent editor History tab with update/create events and Restore actions.
Use the History tab to restore if a limit cuts off valid work.

A stricter stop policy can hide valid multi-step work. If users report early stops, restore the previous version or raise the bound and rerun the same eval.

API references for automation

Section titled “API references for automation”