chore: condense the output of preflight runs

[mcncl]

Description

preflight runs will consume a lot of tokens with the current way we report to the LLM, this attempts to fix that by reducing the information density.

Changes

• in JSON output, only returns a payload which contains relevant information for an LLM to use in order to get information on which jobs have failed

Caveats

This is entirely open to change/improvement

Testing

• Tests have run locally (with go test ./...)
• Code is formatted (with go fmt ./...)

Example build_summary JSON now looks more like:

  {
    "type": "preflight_summary",
    "timestamp": "2026-04-08T12:49:28.246592+10:00",
    "preflight_id": "019d6afe-407e-742a-a532-9eaf247099c7",
    "pipeline": "buildkite/cli",
    "build_number": 185541,
    "build_url": "https://buildkite.com/buildkite/cli/builds/185541",
    "build_state": "failed",
    "failed_jobs": [
      {
        "id": "019d6afe-80b8-4f62-a0e4-8871e94b977e",
        "name": ":pipeline: Upload Pipeline",
        "command": "./.buildkite/pipeline.sh | buildkite-agent pipeline upload",
        "state": "failed",
        "exit_status": 128,
        "soft_failed": false,
        "retried": false
      }
    ],
    "duration_ns": 27448501500
  }

For a live job_failure event:

  {
    "type": "job_failure",
    "timestamp": "2026-04-08T12:49:27.390000+10:00",
    "preflight_id": "019d6afe-407e-742a-a532-9eaf247099c7",
    "pipeline": "buildkite/cli",
    "build_number": 185541,
    "build_url": "https://buildkite.com/buildkite/cli/builds/185541",
    "job": {
      "id": "019d6afe-80b8-4f62-a0e4-8871e94b977e",
      "name": ":pipeline: Upload Pipeline",
      "command": "./.buildkite/pipeline.sh | buildkite-agent pipeline upload",
      "state": "failed",
      "exit_status": 128,
      "soft_failed": false,
      "retried": false
    }
  }

Disclosures / Credits

Written by Codex. YOLO acceped by me.