kaxil opened a new pull request, #74737: URL: https://github.com/apache/airflow/pull/74737
The Common AI approval-gates guide explained every parameter but never showed the reviewer's side: where a pending review appears, what the form contains, and what each answer does to the task and to the tasks after it. This adds that walkthrough to `approval_gates.rst`, built around a complete Dag a reader can copy: an `LLMOperator` with `require_approval=True` and `allow_modifications=True`, its input in the snippet, and a downstream `publish` task that receives the approved text.  The page shows the Required Actions list and the review form, then a table of what approve, edit-then-approve, reject and the three `on_approval_timeout` settings do: the resulting task state, the line the task logs when it resumes, and whether `publish` runs and with what text. Every state, log line and screenshot comes from a run on main (breeze, LocalExecutor). The connection pointed at a local stand-in for the OpenAI API so no key was needed, and the page says a real model's wording will differ. The Dag is a new `example_llm_approval.py` instead of wider markers in `example_llm.py`, because that file's shared import block also pulls in `UsageLimits`, `Decimal` and `BaseModel`, so no marker range there gives a copyable Dag with its imports. The existing notifier example in `example_llm.py` stays where it was, now under a Timeouts heading. Running it also turned up a version split the docs didn't mention. Since #73235 (3.4.0b1), a retried or cleared task rewrites the open review: subject, body, editable output and assigned users. On 3.1 to 3.3 the review keeps the first attempt's, so with `allow_modifications=True`, approving it unedited returns the first attempt's output. The guide, the `operators/llm.rst` parameter list and the `LLMOperator` and `LLMApprovalMixin` docstrings stated the older behaviour for every version; they now say which version does what. -- 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]
