mirror of
https://github.com/BerriAI/litellm.git
synced 2026-09-10 22:41:41 +00:00
7064b2bf09
The workflow now triggers on pull_request instead of pull_request_target, which the auto-approve one-pager and the 2026-06-18 CI exposure post-mortem both forbid, and the job skips fork PRs since the read-only token cannot label them. The floor now follows the rubric's author row (Devin-opened low, human-opened medium, fork high) through a new authors.low list, and adds a paths.medium list for agent instruction files, conftest, helm, and docker. Two guards compare both sides of a file through git show: model map rows that change or disappear stop being low, and a budget file that raises a limit stops being low. A skip or importorskip call added to an existing test file with no net new test scores high. The last 400 staging merges score 218 high, 179 medium, 3 low; the author row alone moves 37 human-opened docs and test PRs from low to medium |
||
|---|---|---|
| .. | ||
| _agent_shin_actions.py | ||
| agent_shin_shared.py | ||
| assert_ci_coverage.py | ||
| assert_workflow_dir_hygiene.py | ||
| auto_update_price_and_context_window_file.py | ||
| close_low_quality_prs.py | ||
| detect_changes.sh | ||
| e2e_egress_sentinel.py | ||
| e2e_fetch_fixture_bundle.sh | ||
| e2e_pack_fixture_bundle.sh | ||
| risk_tier.py | ||
| run_llm_translation_tests.py | ||
| scan_keywords.py | ||
| select_ui_test_scope.sh | ||
| smoke_test_native_wheel.py | ||
| triage-requirements.txt | ||
| triage_with_llm.py | ||
| uv_sync_with_retries.sh | ||
| verify_linux_native_wheel.py | ||