Examples

Five worked examples. Each has a concrete schema, a query pattern, and the actions it drives. Copy a schema and adapt it for your workflow.

Multi-agent append-heavy brainstorming. Ideas with ICE scores sorted and filtered through CLI. Exports to Miro or a report.

Partial candidates accumulate through split/dedupe/estimate. Commit passes fully-formed entries to Jira; logbook gets ticket_id patched back.

One row per observed failure mode per run. Append-only with correction rows. Severity patterns surface only across many runs.

Multi-table logbook: eval definitions, per-run results, iteration aggregates, human feedback — joined on eval_id, run_id, iteration_id.

One row per meaningful step: phase, action, status, artifact_ref, next_intent, blocked_on. Supervisors query without reading transcripts.