Choose Output Artifacts and Logs¶
Choose an output from its intended side effect. File exporters create Platform artifacts; LogExporter writes a small diagnostic projection to the task log. Console output is a runtime implementation aid and is deliberately not offered as an Agent Authoring target.
| artifact-and-log-targets.xml | |
|---|---|
1 2 3 4 5 6 7 8 | |
This creates a semicolon-delimited CSV artifact, an NDJSON artifact, and three bounded diagnostic log entries from the same product. Use multiple targets only when every side effect is an explicit requirement.
Agent Authoring boundary¶
Authoring keeps these intents separate:
file_exportrequires a registered artifact format such asCSV,JSON,XML,XLSX,TXT,PARQUET,SQL,Template, orOpenSearchBulk.logrenders the registeredLogExporterand has no format field.memstore, database, MongoDB, Kafka, and RabbitMQ are distinct target families because they have different lifetime, client, operation, and delivery contracts.
The live Authoring catalog supplies the exact registered file-format set. Agents must not invent an exporter name or represent logging as a file format.
File-target choices¶
- Use
CSVfor tabular interchange and set options such asdelimiter,quotechar,quoting, andline_terminatoronly when the consumer requires them. - Use
JSON(use_ndjson=True)for one JSON object per line; bareJSONwrites an array. - Use
JSONSingle,XMLSingle, orTemplateSinglewhen every record needs its own artifact. These routing-capable exporters may use dynamictargetEntity; streaming multi-record exporters may not. - Use
Templatefor a project-ownedname.ext.templatefile when the target format is not otherwise modeled. Missing-placeholder behavior and record separators are explicit target options. - Use
OpenSearchBulkfor OpenSearch bulk-import payloads. Metadata fields are$$_action$$,$$_index$$,$$_id$$, and$$_routing$$; other fields form the document body.
On DATAMIMIC Platform, omit storageId and exportUri so generated files use the task-scoped artifact location shown in Artifact View. Set exportUri only with a storageId that selects your own <object-storage>; it is a relative path prefix, not a filename, and those outputs are outside Platform Artifact View. The generated <generate> reference owns the current attributes and rules; the Authoring target reference owns agent-facing intent models.