How should an AI SRE end its run when high latency turns out to be nothing?

Tares lets an AI SRE end that run with the conclude tool and the no_op outcome, so no finding is recorded and nothing lands on the project’s Overview. If you want a record that high latency was noise, conclude with a finding instead, which carries a verdict, a headline and a next step. Turn on the agent’s conclusion so the run always ends on purpose, whatever its prompt says.

How it works

An agent set to end with a conclusion gets the conclude tool. The tool offers two ways to finish a run:

• no_op: there is nothing to report. No finding is recorded. When a latency alert fires and the agent finds normal load, a finished deploy or a brief spike that is already over, this is the clean exit.
• A finding with a verdict about a named entity. A finding also carries a headline and a next_step.

The two extra fields have fixed limits:

• headline: one line of up to 100 characters, what was found in the reader’s terms.
• next_step: up to 300 characters, what a person should do. It is empty when nothing needs doing.

The project’s Overview shows the headline and next step on each card, and agents are told to give both. If a run has none, Tares takes the headline from the note’s heading, bold lead or first sentence. It takes the next step from the text after “Next step:”, “Next:” or “Recommendation:”.

The choice between the two for a benign latency run is yours. Use no_op when you do not want anything recorded. Use a finding with a verdict when you want the project to show that the agent looked and found it was noise. The Root cause after a handoff preset comes with a resolved verdict for exactly that case: it was noise, or it is already over. The Triage (on a schedule) preset comes with investigate.

The agent’s summary shows how it ends. Its handoffs key off the verdict words, so a verdict like resolved can be used by whatever you chain after the agent. See How a run ends for the full behavior.

Set it up

1. Open the agent in the agent form and go to the How it ends section.
2. Tick Ends every run with a conclusion. In config this is concludes: true. The agent then always gets the conclude tool, and Tares adds the instructions for it, so your prompt only describes the job. An agent whose prompt already names conclude still gets the tool. Ticking the box makes it required.
3. Under Verdicts it can give, add your own words, each with when to give it. You can add up to 10. Each is one word made of letters, digits or dashes, each is listed once, and none can be no_op, because nothing to report is always possible.
4. In the prompt, describe the job and say what counts as nothing for your service, for example latency that is back to normal. Say that the agent should conclude with no_op in that case.
5. Save the agent and let it run from its trigger. This is the shape of the config:

agents:
  - name: triage
    trigger: watch-web
    concludes: true
    verdicts:
      - { verdict: investigate, when: one entity looks like a problem and needs a closer look }
    prompt: |
      ...

With a verdict list in place, the tool accepts only those words and tells the model what each one means. A finding must carry one. If the agent gives a word that is not on the list, or leaves the verdict out, the call goes back to the agent to fix within the same run. no_op is not on that list and is always accepted. Without a list, a finding can carry any verdict, or none.

This setup is one part of the wider AI SRE workflow, where agents watch, judge and hand off.

What you get

• Runs that end on purpose. If an agent stops without concluding, Tares asks once more with conclude as the only choice.
• A quiet exit for false alarms. A no_op run records no finding, so the Overview is not filled with cards that say nothing happened.
• A readable card when you do record a finding. It has a one-line headline of up to 100 characters and a next step of up to 300 characters, so a person can see at a glance whether anything needs doing.
• Consistent vocabulary. With a verdict list, every finding uses words you chose, and each word has a stated meaning the model sees.
• Clean handoffs. Because handoffs key off verdict words, a downstream agent can act on investigate and ignore resolved.

When something else fits

• If you leave the conclusion off, the agent’s last message is its finding and it has no verdict. That suits a free-form note, but it gives you no no_op exit and nothing for handoffs to key off.
• If you want a durable record that a latency alert was reviewed and dismissed, conclude with a finding and a verdict such as resolved instead of no_op.

Related