Skip to content

Add dedicated BC PR Review evaluation runner - #800

Open
Wenjie Fan (gggdttt) wants to merge 3 commits into
revert-774-bench/engine-review-runnerfrom
gggdttt-merge-pr-795
Open

Add dedicated BC PR Review evaluation runner#800
Wenjie Fan (gggdttt) wants to merge 3 commits into
revert-774-bench/engine-review-runnerfrom
gggdttt-merge-pr-795

Conversation

@gggdttt

@gggdttt Wenjie Fan (gggdttt) commented Aug 17, 2026

Copy link
Copy Markdown
Collaborator

Summary

Reapply the decoupled code-review workflow on top of revert-774-bench/engine-review-runner so the complete feature can be reviewed as one focused diff.

  • Keep code-review available to the generic Copilot and Claude runners.
  • Add a dedicated pr-review runner backed by the production BC PR Review engine and BCQuality.
  • Run the engine's complete local, non-posting pipeline and consume its normalized al-code-review-findings.json artifact.
  • Reject failed, unknown, and malformed engine outcomes instead of scoring them as clean reviews.
  • Collect token usage, Premium Requests/AI Credits, and exact engine/filtered-BCQuality provenance.
  • Pin BC-ALAgents to f2ac8704bf8d39000f8002bcf2d287f1f5b5e9ba, default to gpt-5.6-luna, simplify the CLI, and bump BC-Bench to 0.9.0.

Dependency

CI authentication for local engine reviews depends on microsoft/BC-ALAgents#57, whose exact commit is pinned by this PR.

Validation

  • pre-commit run --all-files
  • pytest: 750 passed, 2 skipped
  • Real local run of synthetic__style-018 with the pinned engine completed through the production non-posting path, emitted normalized output, and parsed 5.9M prompt tokens, 51K completion tokens, and 50.6 AI Credits.

Context

The target branch reverts #774. Using it as the base makes this PR show the full feature rather than only the follow-up decoupling changes from #795.

Keep code-review available to generic Copilot and Claude runners while adding a production-fidelity runner backed by the BC PR Review engine.

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
@gggdttt Wenjie Fan (gggdttt) changed the title Decouple code-review from the BC PR Review runner Add dedicated BC PR Review evaluation runner Aug 17, 2026
Keep the workflow dispatch choice aligned with the Copilot CLI model values introduced by #796.

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

@haoranpb Sun Haoran (haoranpb) left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Try to test by running this locally, since it's a brand new path

Comment thread .github/workflows/pr-review-evaluation.yml Outdated
Comment thread docs/code-review.md Outdated
Comment thread src/bcbench/agent/pr_review/agent.py Outdated
Comment thread src/bcbench/agent/pr_review/agent.py Outdated
Comment thread src/bcbench/agent/pr_review/agent.py
Comment thread src/bcbench/commands/evaluate.py Outdated
Comment thread src/bcbench/commands/evaluate.py Outdated
Comment thread src/bcbench/commands/run.py Outdated
Comment thread src/bcbench/cli_options.py
Comment thread src/bcbench/types.py Outdated
Pin the engine, run its full non-posting production pipeline, collect cost and provenance metrics, and simplify the runner configuration and CLI.

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

Copilot-Session: 003aa3eb-6da6-49f0-bb28-49c8bb0e0d3f
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants