Skip to content

Add agentic review for management AutoPRs - #50039

Open
Weidong Xu (weidongxu-microsoft) wants to merge 14 commits into
mainfrom
weidongxu/management-autopr-agentic-review
Open

Add agentic review for management AutoPRs#50039
Weidong Xu (weidongxu-microsoft) wants to merge 14 commits into
mainfrom
weidongxu/management-autopr-agentic-review

Conversation

@weidongxu-microsoft

Copy link
Copy Markdown
Member

Summary

  • add an agentic workflow for reviewing generated Java management AutoPRs
  • add focused reviewer and critic agents with management-specific safeguards
  • add Vally evaluation coverage for findings, orchestration, and true negatives

Validation

  • vally lint .github/skills/management-autopr-review

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

Copilot-Session: 8cab2a2a-1ca4-44ea-8836-2c9b6dc2cfa9
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot-Session: 8cab2a2a-1ca4-44ea-8836-2c9b6dc2cfa9
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot-Session: 8cab2a2a-1ca4-44ea-8836-2c9b6dc2cfa9
Embed the critic contract in the inline sub-agent and runtime-import trusted reviewer rules into the main prompt so branch trials do not depend on files already existing on Azure main.

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

Copilot-Session: 8cab2a2a-1ca4-44ea-8836-2c9b6dc2cfa9
Use GA CHANGELOG breaking entries as the primary evidence, ignore beta breaks and generated paths, and derive folder identity from the third module-name segment.

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

Copilot-Session: 8cab2a2a-1ca4-44ea-8836-2c9b6dc2cfa9
Treat a missing release-plan link as a nonblocking PR-description concern so GA breaking changes remain visible.

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

Copilot-Session: 8cab2a2a-1ca4-44ea-8836-2c9b6dc2cfa9
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
@azure-pipelines

Copy link
Copy Markdown
Azure Pipelines:
Successfully started running 1 pipeline(s).
34 pipeline(s) were filtered out due to trigger conditions.
There may be pipelines that require an authorized user to comment /azp run to run.

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

Weidong Xu (weidongxu-microsoft) commented Aug 5, 2026

Copy link
Copy Markdown
Member Author

We've got a few month's review on my local agent, and we've collected some typical pattern to raise. Now add a agentic workflow.

We will fix/enhance it as we see more cases.

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

Weidong Xu (weidongxu-microsoft) commented Aug 5, 2026

Copy link
Copy Markdown
Member Author

Validation completed on the current branch:

  • gh aw compile management-autopr-review --approve --actionlint succeeds with no warnings.
  • Vally findings: 100% across 14 stimuli, including all three MGMT-MANAGER-NAME signals and severity assertions.
  • Vally orchestration: 100% across 3 stimuli × 3 trials, including critic dispatch and prompt-injection resistance.
  • Vally true negatives: 100% across 8 stimuli × 3 trials, including valid/reordered manager names, unchanged legacy names, generated-only gating, and prior-concern deduplication.
  • CSpell reports 0 issues for the reviewer Markdown changes.
  • The manager-name implementation was pushed in 7709c069bf0.

@weidongxu-microsoft
Weidong Xu (weidongxu-microsoft) marked this pull request as ready for review August 5, 2026 15:55
Copilot AI lite review requested due to automatic review settings August 5, 2026 15:55
@azure-pipelines

Copy link
Copy Markdown
Azure Pipelines:
Successfully started running 1 pipeline(s).
34 pipeline(s) were filtered out due to trigger conditions.
There may be pipelines that require an authorized user to comment /azp run to run.

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

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

Pull request overview

Adds a GitHub Agentic Workflows-based, read-only “management AutoPR” reviewer for Java management-library generated PRs, including a dedicated reviewer persona, a false-positive “critic” sub-agent protocol, and a Vally evaluation suite to validate findings/orchestration/true-negatives behavior.

Changes:

  • Introduces a pull_request_target workflow (plus compiled lock) that gates on eligible AutoPRs and posts at most one advisory comment via safe-outputs.
  • Adds management-specific review rules (folder/version/LRO/api-version overlap/breaking/release-plan) and a critic protocol for false-positive filtering.
  • Adds a Vally eval suite with fixtures to validate detection, orchestration, and true-negative behavior; updates cspell dictionary for new terms.

Reviewed changes

Copilot reviewed 27 out of 28 changed files in this pull request and generated 3 comments.

Show a summary per file
File Description
.vscode/cspell.json Adds AutoPR/eval fixture vocabulary to avoid spellcheck noise.
.github/workflows/management-autopr-review.md Defines the management AutoPR review workflow prompt, gates, safe-outputs, and critic dispatch contract.
.github/workflows/management-autopr-review.lock.yml Compiled gh-aw workflow used by GitHub Actions to run the reviewer.
.github/skills/management-autopr-review/SKILL.md Management-specific review rules and output contract for the unattended reviewer.
.github/skills/evals/management-autopr-reviewer/vally/eval-true-negatives.yaml True-negative suite ensuring low-noise behavior and deduplication.
.github/skills/evals/management-autopr-reviewer/vally/eval-orchestration.yaml Orchestration tests for injection resistance and critic usage.
.github/skills/evals/management-autopr-reviewer/vally/eval-findings.yaml Positive detection coverage for each supported concern type.
.github/skills/evals/management-autopr-reviewer/run-evals.ps1 Helper script to run Vally suites locally and emit junit output.
.github/skills/evals/management-autopr-reviewer/README.md Documents the eval intent, constraints, and how to run suites.
.github/skills/evals/management-autopr-reviewer/fixtures/tn-service-folder-match.md Fixture for service-folder match true-negative.
.github/skills/evals/management-autopr-reviewer/fixtures/tn-prompt-injection.md Fixture for prompt-injection resistance true-negative.
.github/skills/evals/management-autopr-reviewer/fixtures/tn-prior-concern.md Fixture for carried-forward concern deduplication behavior.
.github/skills/evals/management-autopr-reviewer/fixtures/tn-generated-only-java.md Fixture ensuring generated-path-only Java changes are ignored.
.github/skills/evals/management-autopr-reviewer/fixtures/tn-beta-breaking-changelog.md Fixture ensuring MGMT-BREAKING is not raised for beta packages.
.github/skills/evals/management-autopr-reviewer/fixtures/tn-additive-beta-update.md Fixture for additive beta update true-negative (no concerns).
.github/skills/evals/management-autopr-reviewer/fixtures/positive-version.md Fixture for MGMT-VERSION detection (stable package on preview API).
.github/skills/evals/management-autopr-reviewer/fixtures/positive-service-folder-mismatch.md Fixture for MGMT-FOLDER detection.
.github/skills/evals/management-autopr-reviewer/fixtures/positive-missing-release-plan.md Fixture for MGMT-RELEASE-PLAN detection.
.github/skills/evals/management-autopr-reviewer/fixtures/positive-lro.md Fixture for MGMT-LRO detection.
.github/skills/evals/management-autopr-reviewer/fixtures/positive-folder.md Fixture for MGMT-FOLDER detection in a “collision” scenario.
.github/skills/evals/management-autopr-reviewer/fixtures/positive-breaking.md Fixture for MGMT-BREAKING detection based on GA changelog section.
.github/skills/evals/management-autopr-reviewer/fixtures/positive-api-version-overlap.md Fixture for MGMT-API-VERSION-OVERLAP detection.
.github/skills/evals/management-autopr-reviewer/.vally.yaml Defines Vally suites (all, true-negatives) and paths.
.github/skills/evals/management-autopr-reviewer/.gitignore Ignores eval results output directory.
.github/aw/actions-lock.json Updates pinned gh-aw setup action version used by compilation.
.github/agents/protocols/management-autopr-review-critic.protocol.md Defines the reviewer→critic dispatch contract and verdict semantics.
.github/agents/management-autopr-reviewer.agent.md Reviewer persona, calibration, gates, and critic-dispatch sequence.
.github/agents/management-autopr-review-critic.agent.md Critic persona instructions for filtering false positives.

Comment thread .github/skills/management-autopr-review/SKILL.md Outdated
Comment thread .github/workflows/management-autopr-review.lock.yml
Comment thread .github/skills/evals/management-autopr-reviewer/README.md Outdated
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
@weidongxu-microsoft

Weidong Xu (weidongxu-microsoft) commented Aug 6, 2026

Copy link
Copy Markdown
Member Author

Real-PR dry runs were completed from the private trial host. Safe outputs were staged, so no Azure PR was modified.

  • PR #50024, GA package: current-rules trial staged MGMT-RELEASE-PLAN and MGMT-BREAKING warnings plus MGMT-API-VERSION under Informational with API version 2026-07-01.
  • PR #49732: current-rules retry correctly returned noop because the PR is now closed. Its earlier eligible beta-package trial returned noop without a false MGMT-BREAKING finding.

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