Imported from joelhooks/joelclaw (skills/o11y-logging/SKILL.md). Install upstream with npx skills add joelhooks/joelclaw --skill o11y-logging. Copyright stays with the author.

JoelClaw Observability + Logging

Prevent silent failure by default. Observability is not optional polish: it is part of done.

Non-Negotiable Rules

1. Use the canonical event contract only.
   • packages/system-bus/src/observability/otel-event.ts
   • packages/system-bus/src/observability/emit.ts
   • packages/system-bus/src/observability/store.ts
2. Worker/Inngest code emits through emitOtelEvent or emitMeasuredOtelEvent.
3. Gateway code emits through emitGatewayOtel.
4. Internal ingestion goes through POST /observability/emit (packages/system-bus/src/serve.ts), not ad-hoc writes.
5. Never treat console.log as primary observability. Keep structured events as source of truth.
6. High-cardinality values go in metadata, not in facet fields (source, component, level, success).
7. Failures must set success: false with a meaningful error.
8. Verify delivery to the sinks required by current telemetry configuration. Inspect mirror enablement first; a retired or optional mirror does not block unrelated work.
9. In Inngest durable functions, any "emit once" telemetry must live inside step.run(...) to avoid replay duplication after resume.

Event Conventions

• source: subsystem (worker, gateway, webhook, memory, verification, etc.)
• component: stable module/service name (check-system-health, redis-channel, observe)
• action: stable dotted action (system.health.checked, events.immediate_telegram)
• metadata: request IDs, deployment IDs, function IDs, session IDs, payload identifiers
• duration_ms: include for timed operations

Use event-per-hop (wide event style): one context-rich event for each major boundary/operation, not scattered string logs.

Implementation Workflow

1. Identify the boundary being changed.
   • Inngest function, gateway channel, webhook route, API route, background job, sync step.
2. Add success and failure envelopes.
   • Start + completion for long tasks, or a single completion event for short tasks.
3. Include operational and business context in metadata.
   • Example: function id, event id, provider, queue depth, affected resource id.
4. Keep severity useful.
   • debug/info for normal activity, warn for degraded but recoverable, error/fatal for failures.
5. Run verification gates before finishing.

For full checklists and command recipes, read references/implementation-checklist.md.

Quick Patterns

Worker / Inngest timed operation

import { emitMeasuredOtelEvent } from "../../observability/emit";

await emitMeasuredOtelEvent(
  {
    level: "info",
    source: "worker",
    component: "content-sync",
    action: "content_sync.run",
    metadata: { trigger: event.name },
  },
  async () => {
    await runSync();
  }
);

Gateway emission

import { emitGatewayOtel } from "../observability";

await emitGatewayOtel({
  level: "error",
  component: "redis-channel",
  action: "events.immediate_telegram",
  success: false,
  error: "telegram_send_failed",
  metadata: { sessionId, queueDepth },
});

CLI emission

Use --metadata for JSON context. There is no --attributes flag.

joelclaw otel emit "task.completed" \
  --source system \
  --component skills \
  --success true \
  --metadata '{"session":"NimbleBadger","task":"install wzrrd-publish skill"}'

Definition of Done

• Structured OTEL events added for the changed path.
• No direct feature-level writes to Typesense/Convex for observability data.
• Smoke probe passes (scripts/otel-smoke.sh).
• joelclaw otel list and joelclaw otel stats show expected behavior.
• New failure modes are queryable by source, component, and action.

Inngest Replay + Hang Triage

Use this when step code appears to run but runs remain RUNNING/CANCELLED with Finalization errors.

1. Inspect run trace first.

joelclaw run <run-id>

Look for errors.Finalization.stack containing Unable to reach SDK URL.

2. Confirm whether this is true network reachability or worker-side blocking.

joelclaw inngest status
joelclaw logs worker --lines 200
joelclaw logs errors --lines 200

3. Check for replay-noise in OTEL.

If an action that should emit once (for example manifest.archive.prereqs-passed) appears hundreds of times in one run window, move that emit into its own step.run.

joelclaw otel search "manifest.archive.prereqs-passed" --hours 1

4. Treat Unable to reach SDK URL as an ambiguous symptom.

It can indicate ingress problems, but in practice it can also happen when a function handler blocks on local IO/dependencies long enough that finalization cannot complete.

Helper Script

Use scripts/otel-smoke.sh for a fast end-to-end probe:

./skills/o11y-logging/scripts/otel-smoke.sh verification o11y-skill probe.emit

Key Files

• packages/system-bus/src/observability/otel-event.ts
• packages/system-bus/src/observability/emit.ts
• packages/system-bus/src/observability/store.ts
• packages/system-bus/src/serve.ts
• packages/gateway/src/observability.ts
• packages/system-bus/src/inngest/functions/check-system-health.ts
• packages/cli/src/commands/otel.ts
• apps/web/app/api/otel/route.ts