Triage an incident step by step

Each answer builds on the ones before it: the severity decides the rollback, and both shape the status message.

depends_on
Enum
Text

Challenges

The answers must agree with each other: whether to roll back depends on how severe it is, and the status message has to say what was decided.
With depends_on the fields are answered in order, each one seeing the answers before it, instead of three separate guesses that could contradict each other.
The status message is free text, and it still has to match the two typed answers before it.

import os
from typellm import TypeLLMClient

client = TypeLLMClient(api_key=os.environ["TYPELLM_API_KEY"])
response = client.generate(
    context="Alert at 14:02: checkout API error rate 38% (normal: under 1%) since the 13:55 deploy of payments-service v2.14. p95 latency 4.1 s. About 1,200 customers affected so far.",
    questions={
      "severity": {
        "type": "string",
        "enum": [
          "low",
          "medium",
          "high",
          "critical"
        ],
        "instructions": "How severe is this incident?"
      },
      "roll_back": {
        "type": "boolean",
        "depends_on": [
          "severity"
        ],
        "instructions": "Given the incident and its severity, should the deploy be rolled back now?"
      },
      "status_message": {
        "type": "string",
        "depends_on": [
          "severity",
          "roll_back"
        ],
        "instructions": "One sentence for the status page, saying what is affected and what is being done."
      }
    },
)
print(response)

Time 1.06 s · Cost $0.000009

Generation(
    result={
        'severity': 'critical',
        'roll_back': True,
        'status_message': 'We are experiencing elevated error rates and latency in the checkout process affecting approximately 1,200 customers, and we are immediately rolling back the recent payments-service deployment to restore service.',
    },
    thinking={},
    usage=Usage(input_tokens=179, thinking_tokens=0),
)