debugging-signals-pipeline

Query the Temporal REST API at localhost:8081 to list recent workflows and fetch workflow history by ID. The history events show which activity failed, including failure messages, timers, and signal names.

Run python manage.py emit_signals_from_fixture with a type like zendesk, a team ID, offset, and limit. Always run cleanup_signals first to avoid stale embeddings causing phantom report matches.

This error occurs when stale embeddings remain in ClickHouse after cleanup because CLICKHOUSE_DATABASE is unset, so cleanup targets the default database instead of posthog. Set CLICKHOUSE_DATABASE=posthog in .env and manually delete stale embedding rows.

Fetch the log_url from the most recent TaskRun model in the Django shell, then read it with object_storage.read. The JSONL log tail reveals agentsh network denials, progress events, and sandbox provisioning messages.

The sandbox Claude agent went idle longer than TASKS_INACTIVITY_TIMEOUT_SECONDS, which defaults to two hours. Set it to 30 seconds locally for fast failures, then check the agent log for agentsh network denials or invalid ANTHROPIC_API_KEY.

No, sandbox containers are ephemeral and removed after the task run completes. Inspect them while running using docker ps with the task-sandbox name filter, docker exec for processes, and reading /tmp/agent-server.log.