xuang7 opened a new issue, #7517: URL: https://github.com/apache/texera/issues/7517
### Feature Summary Texera has ~160 operators, but most of them currently do not have demo videos. Recording and maintaining a separate demo manually for every operator does not scale. This proposal introduces a Playwright-based tool that automatically generates a short demo video for each operator, showing how the operator is added to a workflow, what values to enter in its form, and what output or behavior to expect. Operator authors need to maintain the sample field values for their operators. Everything after that would be generated deterministically by shared automation maintained in one central place. The videos would be regenerated as part of each major release and hosted online. As an additional benefit, the generation process exercises the full operator workflow end to end: logging in, creating a workflow, dragging an operator onto the canvas, filling in its properties, running the workflow, and viewing the results. This also provides end-to-end coverage for common operator interactions. ### Proposed Solution or Design A Playwright-based generator that produces one short demo video per operator, driven by agent-generated sample values that are reviewed by the operator author. 1. Agent-generated sample values, reviewed by the author. Each operator has an entry in `operator-field-values.json` describing the values to populate in its form. These values are generated by an agent using the operator schema together with the sample dataset, so fields that depend on dataset columns can be populated with valid examples. The operator author reviews the generated values before they are committed. A validator checks each entry against the operator's actual schema and the sample dataset so that unknown fields, invalid value types, and references to non-existent columns fail early. 2. Deterministic script generation. `OperatorScriptGenerator` reads the operator metadata and reviewed sample values, selects the appropriate workflow template and wiring strategy based on the operator group and port topology (for example, source, single-input, or multi-input), and emits a per-operator script. The same inputs always produce the same script, and no LLM is involved at this stage. 3. Recorded execution. Shared controllers drive a real browser through the demo flow: opening the prepared workflow, dragging the operator onto the canvas, filling in its properties, running the workflow, and opening the result panel. Recording begins after the sample workflow is imported. 4. Publishing. Generated videos are written locally. A maintainer publishes them externally, for example to YouTube. Regeneration is performed as part of each major release. Architecture diagram: <img width="623" height="533" alt="architecture" src="https://github.com/user-attachments/assets/36b91763-8273-48f4-9f3c-40f0f9a7475b"; /> Sample video: [line-chart_demo.webm](https://github.com/user-attachments/assets/4e184a50-0f68-4beb-a5bb-25f63e7b7681) This proposal adds a new `Docs` sbt module with a Playwright dependency. No LLM code or API keys are included in the repository. The agent is only used during the offline sample-value generation step, while the reviewed configuration is the only generated input committed to the repository. #### Future improvements A possible future extension is to use an LLM-assisted repair loop when generation or execution fails, especially since the current implementation relies heavily on UI interactions. For example, the agent could inspect validation errors or failed Playwright runs and suggest or automatically generate corrected sample values or interaction steps. This would remain separate from the deterministic generation path described above and could be explored once the initial workflow is stable. Discussion: #6871 ### Affected Area _No response_ -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
