{
  "slug": "vercel-ci-logs-displays-and-follows-task-logs-for-a-run-3caf8a3b32db",
  "title": "vercel ci logs displays and follows task logs for a run",
  "tools": [
    "vercel"
  ],
  "tags": [
    "agent"
  ],
  "introduced_at": "2026-10-01T04:15:23.555Z",
  "announced_at": "2026-10-01T04:15:23.555Z",
  "effective_at": null,
  "date_basis": "source-publication",
  "status": "unknown",
  "how": {
    "incantation": "When debugging a Vercel CI run, use `vercel ci logs <run-id>` to inspect task logs; add the documented filters or `--follow` to stream a running task. If the run fails before any tasks start, retrieve its setup logs.",
    "examples": [
      "vercel ci logs <run-id>",
      "vercel ci logs <run-id> --failed --task <task> --level <level>",
      "vercel ci logs <run-id> --follow"
    ],
    "gotchas": [
      "Version: 62.1.0",
      "<run-id> (required): Vercel CI run to inspect.",
      "--failed: Filter task logs to failed tasks.",
      "--task <task>: Filter logs by task.",
      "--level <level>: Filter logs by level.",
      "--follow: Stream logs for a running task.",
      "Follow mode performs a final full-snapshot read without the stream timestamp cursor when a task completes, reconnects healthy stream timeouts, handles JSON snapshot responses, respects stdout backpressure, preserves JSONL errors, and bounds interrupted or invalid response retries.",
      "For completed retried tasks, empty logs produce a stderr warning when no level filter is applied; valid quiet, cached, or skipped tasks do not fail. JSON mode emits a structured stderr warning and reserves stdout for log records.",
      "Human output distinguishes runs without failures yet from completed runs; empty `--failed` JSON results include `run.finished`.",
      "Commands resolve the latest attempt from the attempts endpoint instead of relying on list-row ordering.",
      "Reconnects replay retained logs and deduplicate by timestamp, level, message, and occurrence count, preserving repeated identical lines and older late arrivals in append-only history.",
      "Replay stops explicitly at its 8 MiB estimated-state or 50,000 distinct-record limit instead of silently evicting records; the final snapshot recovers retained logs suppressed within the active stream.",
      "Without server record IDs, an ingestion cursor, and a completion watermark, logs ingested after the final snapshot are not guaranteed; the final read drains logs available at that point, not future ingestion."
    ]
  },
  "why_it_matters": "Inspect or follow task logs while debugging a Vercel CI run.",
  "sources": [
    "https://vercel.com/docs/cli/release-notes#vercel@62.1.0"
  ],
  "rank": 0.9904,
  "judge": {
    "publish": true,
    "judge_model": "codex-cli/gpt-6-luna+typesafe/jev-1.13.0",
    "prompt_version": "extract-v7+enrich-v1+judge-v2x3+jev-publish-v1"
  },
  "review_state": "approved",
  "provenance": {
    "run_id": "vercel-2026-10-08",
    "verdict_id": "vercel-2026-10-08:vercel-2026-10-08-a3f2e22fdd5d6a5a-2:gate",
    "evidence_hash": "4d887b6461ad5b58e692176ca723e567928245b46f74342e638f369af61bfbd2"
  },
  "review": {
    "draft_hash": "ad1ea9eed6fb084c19d247dcda748b169512e7aa66b22d59dffcbe137ec4df8d",
    "reviewer": "publication-gate",
    "reviewed_at": "2026-10-07T00:00:00.000Z",
    "decision": "vercel-2026-10-08:gate",
    "reason": "3 of 3 judge runs approved with p_yes >= 0.9 (mean 0.99 >= 0.9; 2 needed) and Jev gave 0.78 (>= 0.6 needed)."
  }
}