jayfarei/opentraces/workflow-authoring

OpenTraces Workflow Authoring

Use this skill when building a workflow for ot dataset run.

Contract

• The dataset schema in schemas/row.schema.json is the only public row shape.
• A workflow emits plain JSONL rows to OT_DATASET_OUTPUT.
• Source traces and raw sessions are immutable inputs. Do not mutate them.
• Use ot trace query first, ot trace slice for dataset-ready windows, ot trace map for bounded verification, and ot trace get only when a bounded packet/slice is insufficient.
• Use ot dataset run <name> --dry-run --limit N --verbose while developing. Dry-runs are never promotable.
• A real run must execute freshly and append only valid non-duplicate rows.

Recommended Loop

1. Read .opentraces/manifest.yaml and schemas/row.schema.json.
2. Run a narrow ot trace query with exact facets or signals.
3. Materialise candidate windows with ot trace slice <trace_id> --template bursts --json.
4. Write or update helper scripts under the workflow package.
5. Emit rows to OT_DATASET_OUTPUT, one JSON object per line.
6. Run the dataset in dry-run mode and inspect validation/dedupe counts.
7. Run the dataset for real only after the dry-run is clean enough.

Example Row

{"source_trace_id":"trace-1","source_unit_id":"tu:trace-1:trace","summary":"The user wanted a stricter design review."}