jobseek: Command for Claude Code

Daily review of jobseek crawler errors. Read-only. Operate from this prompt alone — do not rely on prior conversation context.

================================================================ MISSION

Scan the last 24h of errors on the crawler Hetzner box, classify against prior reviews, write a dated Markdown report, and file a GitHub issue for anything novel, regressing, or spiking. Model output after prior issues #2622, #2621, #2470, #2431. Never mutate remote state.

================================================================ TARGET

Prefer the root-collected runner bundle. For a local manual fallback, resolve the production crawler host from approved operator configuration and never copy its address into reports, traces, prompts, or GitHub content. Compose: /home/deploy/docker-compose.yml

Long-running services — collect logs with docker logs --since <ISO> --until <ISO> <name>: deploy-worker-1-1 HTTP worker deploy-worker-2-1 HTTP worker deploy-worker-3-1 HTTP worker deploy-browser-1-1 Playwright/Lightpanda worker deploy-exporter-1 Postgres → Supabase + Typesense CDC deploy-drain-1 R2 description uploader deploy-indexnow-1 IndexNow notifier deploy-redis-1 local queue (only OOM + level=error lines) deploy-alloy-1 log/metric shipper (only Alloy's own level=error)

Ephemeral one-shots — enumerate with docker ps -a --filter status=exited --format '{{.Names}} {{.Image}} {{.CreatedAt}} {{.Status}}' filtered to containers whose CreatedAt falls inside the window and whose image matches ghcr.io/*/jobseek-crawler:*. These are crawler refresh-typesense | backfill-typesense | sync | notify-indexnow | reconcile runs. Grab their logs with docker logs <id> before they age out (Docker retains exited container logs until the container is pruned).

Ignore interactive debug containers (user-spawned, noisy): names like tesla-debug, stupefied_hofstadter, goofy_haibt.

================================================================ INPUTS

1. Window: last 24h ending at UTC now, rounded down to the minute. Use an explicit --since/--until pair so the window matches the report header exactly. If a runner evidence bundle is available, read manifest.json, host/docker-inspect-state.txt, host/docker-cgroup-memory.json, and host/docker-lifecycle.jsonl first. Also read host/maintenance-correlation.json before classifying a synchronized service stop. The lifecycle stream durably records allowlisted create/start/restart/die/oom/kill/stop/destroy events, exit codes/signals, and pseudonymous container generations; the cgroup file contains generation-safe memory.events counters. Always read host/redis-capacity.prom and host/redis-capacity-observer.log. If manifest.json records redis_capacity.complete=false, report the capacity evidence gap rather than treating the Redis container's cgroup percentage as healthy capacity. Always read the disk_capacity manifest entry and the six bounded files host/docker-system-df.txt, host/docker-container-sizes.txt, host/docker-images.txt, host/disk-attribution.txt, host/docker-gc-timer.txt, and host/docker-gc.log. Treat disk_capacity.complete=false as an attribution gap. When / or Docker is at least 85% full, identify the dominant allowlisted root/image family and check the exact-window GC journal; never close or downgrade the incident from the percentage alone. If evidence is incomplete, name the exact missing artifact instead. Treat active(running) as a healthy in-flight timer and inspect its target service separately. A quiescent timer must be active(waiting) with a finite next trigger; active(elapsed), any other state, or waiting with a missing/infinite next trigger is a failed recurrence guard even while disk usage is safe.
2. Prior review reports: read every .md under ~/dev/claude/review-jobseek-errors/ before classifying. That directory is the agent's cross-run memory.
3. Prior GitHub issues (dedup + pattern matching): gh issue list --label daily-error-review --state all --limit 200
   --json number,title,state,createdAt,body
4. Recently merged PRs (regression correlation — prior issues reference specific PRs like pre-#2232): gh pr list --state merged --limit 40 --json number,title,mergedAt