potiuk opened a new pull request, #68653:
URL: https://github.com/apache/airflow/pull/68653
The `prepare-providers-documentation` skill (reframed in #68641 as an
optional
aid) classified provider changes by spawning **one `Explore` sub-agent per
PR**. On a normal release wave — dozens of providers, hundreds of commits —
that is hundreds of sub-agents, and it's the main reason the run is so long
and
token-heavy.
This replaces the per-PR fan-out with a volume-aware strategy:
- **Few commits remain (≲15 after the cheap auto-classify heuristics) →
classify inline**, no sub-agents at all.
- **More than that → one sub-agent per provider**, classifying that
provider's
entire remaining commit list in a single pass (chunking only providers with
>25 commits).
Multi-provider PRs are already classified independently per provider, so the
provider is the natural batch unit — and one provider-scoped agent amortizes
the breaking-change checklist across all of its commits instead of paying a
fresh agent spin-up per commit. Net effect: sub-agent count drops from
O(commits) to roughly O(providers), and often to zero for incremental
updates.
No behavior change to the classification logic, the breaking-change
checklist,
escalation rules, or the breeze commands — only how the work is fanned out.
---
##### Was generative AI tooling used to co-author this PR?
- [X] Yes — Claude Code (Opus 4.8)
Generated-by: Claude Code (Opus 4.8) following [the
guidelines](https://github.com/apache/airflow/blob/main/contributing-docs/05_pull_requests.rst#gen-ai-assisted-contributions)
--
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]
