Skip to content

Add public API governance instruction evaluation - #68698

Draft
PureWeen wants to merge 6 commits into
mainfrom
pureween-public-api-instruction-eval
Draft

Add public API governance instruction evaluation#68698
PureWeen wants to merge 6 commits into
mainfrom
pureween-public-api-instruction-eval

Conversation

@PureWeen

@PureWeen PureWeen commented Aug 21, 2026

Copy link
Copy Markdown
Member

Summary

  • adds a specialized Vally regression suite for the repository-wide public API governance instructions introduced by Add public API governance instructions #68168 and corrected by Clarify API review timing and add authoring skill #68959
  • verifies that implementation may proceed before API approval, while opening an implementation PR activates the api-review skill and creates a separate linked proposal issue
  • verifies that api-approved covers the final implemented shape before RTM inclusion and that changed approved shapes return to API review
  • loads .github/copilot-instructions.md through environment.files and the complete .github/skills/api-review directory through environment.skills
  • uses GPT-5.6 Luna for both the evaluated agent and judge; this suite does not use Anthropic models

This is the remaining open layer of native stack #68699 after #68168 merged. It follows the specialized-suite conventions under eng/skill-evals and adds no Vally onboarding infrastructure.

Validation

  • pwsh -NoProfile -File ./eng/skill-evals/run.ps1 Test
  • pwsh -NoProfile -File ./eng/skill-evals/run.ps1 Validate -Eval eng/skill-evals/copilot-instructions/public-api-governance.vally.yaml
  • one GPT-5.6 Luna trial per stimulus: 100% overall, all 18 graders passed
    • the implementation-PR scenario activated the api-review skill
    • all four governance lifecycle scenarios passed every deterministic and judge grader
  • git diff --check

@PureWeen
PureWeen changed the base branch from main to pureween-public-api-instructions August 21, 2026 16:38
Base automatically changed from pureween-public-api-instructions to main August 25, 2026 15:52
@PureWeen
PureWeen force-pushed the pureween-public-api-instruction-eval branch from c80cac4 to ba87f27 Compare August 25, 2026 15:52
@Youssef1313 Youssef1313 added the area-infrastructure Includes: MSBuild projects/targets, build scripts, CI, Installers and shared framework label Sep 2, 2026
@dotnet-policy-service

Copy link
Copy Markdown
Contributor

Hey @dotnet/aspnet-build, looks like this PR is something you want to take a look at.

Copilot AI added 2 commits September 7, 2026 08:30
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Align the specialized suite with the RTM approval boundary and exercise the api-review skill handoff introduced by #68959.

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

Copilot-Session: 17c247c0-c8ff-497c-82ac-ef81dd1ba327
@PureWeen
PureWeen force-pushed the pureween-public-api-instruction-eval branch from ba87f27 to 27622b9 Compare September 7, 2026 13:32
Copilot AI added 4 commits September 7, 2026 11:50
Set both the evaluated agent and judge defaults to GPT-5.4 so the specialized suite does not use Anthropic models.

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

Copilot-Session: 17c247c0-c8ff-497c-82ac-ef81dd1ba327
Strengthen the specialized graders so responses must name the api-review skill, proposal labels, PublicAPI.Unshipped distinction, and revised ref-assembly proposal workflow.

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

Copilot-Session: 17c247c0-c8ff-497c-82ac-ef81dd1ba327
Set both the evaluated agent and judge defaults to GPT-5.6 Luna for the specialized public API governance suite.

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

Copilot-Session: 17c247c0-c8ff-497c-82ac-ef81dd1ba327
Load the complete api-review skill through Vally's environment.skills, require its activation for PR authoring, and remove brittle proximity matching from otherwise complete responses.

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

Copilot-Session: 17c247c0-c8ff-497c-82ac-ef81dd1ba327
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

area-infrastructure Includes: MSBuild projects/targets, build scripts, CI, Installers and shared framework

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants