arthur-ai/arthur-engine/arthur-onboard-verify

Arthur Onboard — Step 7: Verify Instrumentation

Read State

cat .arthur-engine.env 2>/dev/null || echo "(no state file)"

Parse ARTHUR_ENGINE_URL, ARTHUR_API_KEY, ARTHUR_TASK_ID.

Tell the User to Run Their Application

Show the required env vars for their platform:

# Mac / Linux:
  export ARTHUR_API_KEY=<ARTHUR_API_KEY>
  export ARTHUR_BASE_URL=<ARTHUR_ENGINE_URL>
  export ARTHUR_TASK_ID=<ARTHUR_TASK_ID>
 
# Windows PowerShell:
  $env:ARTHUR_API_KEY  = "<ARTHUR_API_KEY>"
  $env:ARTHUR_BASE_URL = "<ARTHUR_ENGINE_URL>"
  $env:ARTHUR_TASK_ID  = "<ARTHUR_TASK_ID>"
 
# Windows CMD:
  set ARTHUR_API_KEY=<ARTHUR_API_KEY>
  set ARTHUR_BASE_URL=<ARTHUR_ENGINE_URL>
  set ARTHUR_TASK_ID=<ARTHUR_TASK_ID>

Then ask the user to run the app with inputs that resemble real production data if possible — for example, actual user queries, realistic documents, or representative prompts from their domain. Explain that the traces captured here will be used in the next step to auto-generate continuous evaluations tailored to their use case, so realistic inputs produce much better eval coverage than "hello world" or synthetic test strings. If they don't have real data handy, ask them to describe typical inputs so you can suggest a few representative examples together.

Poll for Traces

Once the user confirms they've run the app, poll for traces (up to 60 seconds):

for i in $(seq 1 20); do
  COUNT=$(curl -s \
    -H "Authorization: Bearer $ARTHUR_API_KEY" \
    "$ARTHUR_ENGINE_URL/api/v1/traces?task_ids=$ARTHUR_TASK_ID&page_size=5" | \
    python3 -c "
import sys, json
d = json.load(sys.stdin)
print(len(d.get('traces') or d.get('data') or []))
" 2>/dev/null || echo "0")
  if [ "$COUNT" -gt "0" ]; then
    echo "TRACES_FOUND=$COUNT"; break
  fi
  echo "No traces yet... attempt $i/20"
  sleep 3
done

Traces found: Confirm success and exit this skill.

No traces after 60s: Provide troubleshooting guidance:

1. Check env vars are set correctly in the app
2. Confirm the app made actual LLM calls during the test run
3. Verify Arthur Engine is running: curl $ARTHUR_ENGINE_URL/health
4. Check the app logs for errors related to OpenTelemetry or Arthur SDK

Offer to retry. This step is non-blocking.