pankajkoti opened a new pull request, #74470:
URL: https://github.com/apache/airflow/pull/74470
The `@task.llm`, `@task.agent`, `@task.llm_sql`, `@task.llm_branch`,
`@task.llm_schema_compare` and `@task.llm_file_analysis` decorators called
`render_template_fields()` after running the decorated function. Airflow
already
renders template fields before `execute()`, so this was a second pass over
every
template field. For `prompt` it was the only pass that saw the text the
function
returned, because `prompt` is still a placeholder during the first one.
As a result, ordinary text containing braces was changed before it reached
the
model. `{{ 100 * 3 }}` arrived as `300`, `{{ var.value.x }}` was resolved,
and an
unbalanced `{% raw %}` failed the task with a `TemplateSyntaxError` on
`prompt`.
For example, a support-ticket triage task fed customer text that quoted a
template
snippet.
The returned text is now sent as is, which is how `@task.llm_batch` already
behaves. Other parameters such as `system_prompt` are still rendered once,
before
`execute()`, and are no longer rendered twice. Anyone who relied on
returning Jinja
from the function should put it in a templated parameter or format the string
inside the function. The docs for each decorator now say so.
---
##### Was generative AI tooling used to co-author this PR?
- [X] Yes (Claude Code, model claude-sonnet-5-5)
Generated-by: Claude Code (claude-sonnet-5-5) following [the
guidelines](https://github.com/apache/airflow/blob/main/contributing-docs/05_pull_requests.rst#gen-ai-assisted-contributions)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
--
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]
