What verdict and next step should an AI SRE record for 5xx errors on an API?

Tares lets an AI SRE end a run with the conclude tool. For 5xx errors on an API, record a finding with a verdict such as investigate, a headline of up to 100 characters, and a next step of up to 300 characters. If nothing needs reporting, end with no_op, which records no finding. The verdict words are yours to define on the agent.

How it works

An agent set to end every run with a conclusion gets the conclude tool, and Tares adds the instructions for it. The agent can end in one of two ways:

• no_op: there is nothing to report, so no finding is recorded.
• A finding: a verdict about a named entity, here the API, plus a headline and a next_step.

For 5xx errors the choice maps like this:

• The API’s 5xx errors look like a real problem and need a closer look. Record a finding. A triage agent uses a verdict such as investigate for one entity that needs attention.
• You found what is failing and why. An agent that runs after a handoff records rca.
• The errors were noise, or are already over. The same agent records resolved.
• Nothing to report. End with no_op.

The two text fields have fixed limits:

• headline is one line of up to 100 characters. It says what was found in the reader’s terms, for example the API returning 5xx errors.
• next_step is up to 300 characters. It says what a person should do, and it stays empty when nothing needs doing.

The project’s Overview shows both on each card, and agents are told to give both. If a run ends without them, 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 verdict words are yours. In the agent form’s How it ends section you can list up to 10 verdicts, each one word made of letters, digits or dashes, listed once, and none of them no_op. Each comes with a note on when to give it. With a list, the tool accepts only those words and tells the model what each means. A finding must carry one, and an off-list word or a missing verdict goes back to the agent to fix within the same run.

Set it up

1. Open the agent that watches your API and go to the How it ends section.
2. Tick Ends every run with a conclusion. This is concludes: true, and the agent then always gets the conclude tool, whatever its prompt says. Your prompt only needs to describe the job, such as looking at the API’s 5xx errors.
3. Add the verdicts the agent can give, each with when to give it. For a triage agent that watches an API, one verdict is enough:

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

4. If you use the Triage (on a schedule) preset, it already comes with investigate. For the agent that picks up after a handoff, use the Root cause after a handoff preset. It comes with rca (you found what is failing and why) and resolved (it was noise, or it is already over).
5. Write the next step as an action for a person. For a run that records investigate, a good pair looks like this:
   • Headline: the API is returning 5xx errors.
   • Next step: what a person should check first.
6. For a run where the 5xx errors have cleared and nothing needs doing, leave the next step empty or end with no_op.

The agent’s summary shows how it ends. Handoffs key off the verdict words, so choose words you want a later agent to react to. See How a run ends for the full list of options.

What you get

• Each run ends on purpose, with either no finding or a finding that names the entity, the verdict, a headline and a next step.
• A card in the project Overview that says what was found about the API and what a person should do, without opening the run.
• A verdict you can trust to be one of your own words, because an off-list word goes back to the agent within the same run.
• A second chance if a run stops without concluding. Tares asks once more with conclude as the only choice.
• Quiet runs stay quiet, because no_op records no finding.

This fits the wider AI SRE flow, where a cheap watching agent flags a problem and a later agent finds the cause.

When something else fits

• If you leave Ends every run with a conclusion off, the agent’s last message is its finding and it carries no verdict. That works for a free-form note, but handoffs then have no verdict word to key off.
• If you do not define a verdict list, a finding can carry any verdict or none. Define the list when you want consistent words across runs.
• If a person only needs the finding in prose, the headline and next step fallback from the note text may be enough.

Related