Skip Vally evaluations on fork pull requests

[Copilot]

The Vally evaluations / eval-job check fails on every pull request originating from a fork: all trials error with Session was not created with authentication info or custom provider.

GitHub does not expose repository secrets to pull_request runs from forks (job log: Secret source: None, and GITHUB_TOKEN limited to read scopes — the requested copilot-requests: write is not granted). COPILOT_GITHUB_TOKEN is therefore empty, the vally copilot-sdk executor cannot create a session, and all 45 trials error out. Fork runs can never pass; only same-repo runs can.

Changes (.github/workflows/vally-eval.yml)

• Added a job-level IS_FORK_PR env expression derived from github.event.pull_request.head.repo.full_name != github.repository.
• Guarded the vally-specific steps — CLI install, build_info.json, server build, eval generation, Run vally, and artifact upload — with if: ${{ env.IS_FORK_PR != 'true' }}. This also avoids ~5 minutes of pointless build time per fork PR.
• Added a Report skipped evaluations step that emits a ::notice:: pointing maintainers to re-run the evaluations from a branch in this repository.

Same-repo pull requests and workflow_dispatch runs are unaffected.

    env:
      IS_FORK_PR: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.repo.full_name != github.repository }}

Note

The evaluations still provide no coverage for fork PRs. If they need to run there, that requires a separate design (e.g. a pull_request_target-based or maintainer-triggered workflow with the usual untrusted-code safeguards) rather than wiring a token into this workflow.

[azure-pipelines]

Azure Pipelines:
Successfully started running 1 pipeline(s).
There may be pipelines that require an authorized user to comment /azp run to run.