What verdict and next step should an AI SRE record for failed deploys?

Tares has an AI SRE agent end a failed deploys run with the conclude tool. If a deploy failed, the agent records a finding with a verdict about the failing entity, a one-line headline and a next step. If nothing failed, it calls no_op and records no finding. You choose the verdict words yourself, and the presets give you investigate, rca and resolved to start from.

How it works

An agent set to end with a conclusion gets the conclude tool. It uses the tool to end a run on purpose, in one of two ways:

• no_op: there is nothing to report, so no finding is recorded. For failed deploys, use it when the deploys in the window all succeeded.
• A finding: a verdict about a named entity, such as the service or repo whose deploy failed. The finding also carries two fields:
  • headline: one line of up to 100 characters, saying what was found in the reader’s terms.
  • next_step: up to 300 characters, saying what a person should do. It is empty when nothing needs doing.

The headline and next step are what the project Overview shows 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 come from your agent. You can list up to 10, each one word (letters, digits, dashes), each listed once, and none of them no_op. With a list, the tool accepts only those words and tells the model what each one 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.

The presets already come with words that fit failed deploys. The Triage (on a schedule) preset comes with investigate. Root cause after a handoff comes with rca (you found what is failing and why) and resolved (it was noise, or it is already over).

Set it up

1. Pick the verdict words for failed deploys. A triage agent that watches deploy events needs one word, investigate, for a deploy that looks like a problem. An agent that takes the case after a handoff needs rca and resolved.
2. In the agent form, open How it ends and tick Ends every run with a conclusion. 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.
3. Add the verdicts, each with when to give it. This example is the triage agent from the preset:

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. In the prompt, describe the job in your own terms, for example to look at recent deploys and name the entity whose deploy failed. You do not need to explain the tool.
5. Check how the agent ends in its summary. If a run stops without concluding, Tares asks once more with conclude as the only choice.

The finding for a failed deploy then reads like this:

• Verdict: investigate at triage, or rca once the cause is found.
• Headline: a single line naming the entity and what failed, within 100 characters.
• Next step: what a person does now, within 300 characters. For a cause that is already fixed, use resolved and leave the next step empty.

For how a run fits into the wider AI SRE flow, see AI SRE. The full reference for the tool and its fields is How a run ends.

What you get

• A clear signal per run. A run ends as either no_op with no finding, or a finding with a verdict. Quiet deploy windows do not fill your Overview.
• Cards you can read at a glance. Each failed deploy shows a headline in the reader’s terms and a next step a person can act on.
• Consistent words. With a verdict list, the agent can only use your words, and a missing or off-list verdict is sent back to be fixed in the same run.
• A fallback when fields are missing. Tares fills in the headline and next step from the note’s text when the agent leaves them out.
• Verdicts you can build on. The agent’s handoffs key off these words, so a triage agent that says investigate can pass the case to a root cause agent.

When something else fits

• If a run needs no verdict, leave Ends every run with a conclusion off. The agent’s last message is then its finding, with no verdict.
• If you do not set a verdict list, a finding can carry any verdict or none. Set a list when you want handoffs or Overview cards to depend on consistent words.

Related