test(ext-agents): guard tier selectors against false-green

[glharper]

Summary

The ext-azure-ai-agents-test workflow selects Tier 0 and Tier 1 functional tests with go test -run <regex>. Go exits successfully when a regex matches no tests, so renaming or adding a Test_AIAgent_ test without updating the tier selectors could silently produce a false-green CI run.

Fix

• Centralize the Tier 0 and Tier 1 regexes as job-level environment variables so validation and execution use the same selectors.
• Anchor both regexes to exact top-level test names.
• Add a Validate tier test selectors step that uses go test -list to discover tests and fails when:
  • no Test_AIAgent_ tests are discovered;
  • either tier matches zero tests;
  • a test is assigned to both tiers; or
  • any discovered test is not assigned to exactly one tier.
• Normalize carriage returns so the check is deterministic when reproduced from Windows Bash as well as on the Ubuntu runner.

The current inventory is validated as 4 Tier 0 tests and 5 Tier 1 tests. The actual test steps reuse the validated regex variables.

Validation

• Ran the exact selector guard locally: it discovered all 9 tests and accepted the current 4/5 split.
• Removed Test_AIAgent_Doctor_Help from the Tier 0 selector in a local negative-control run; the guard failed and identified the uncovered test.
• git diff --check passes.

Fixes #8778

[github-actions]

📋 Prioritization Note

Thanks for the contribution! The linked issue isn't in the current milestone yet.
Thank you for logging this issue; our team is reviewing it. If you need urgent prioritization, tag @RickWinter and @kristenwomack to let us know.

[Copilot]

Pull request overview

Adds CI validation to prevent false-green Azure AI Agents tier test runs.

Changes:

• Centralizes and anchors Tier 0/1 selectors.
• Verifies all nine tests belong to exactly one tier.
• Reuses validated selectors during test execution.

[jongio]

One non-blocking suggestion on the new validation step, inline. The guard logic itself checks out: I ran it against the current package and the two tier regexes select exactly the 4 Tier 0 and 5 Tier 1 tests with full coverage and no overlap.

[jongio]

This guard only needs Go and the test sources to compile. go test -list against ./test/functional/ai_agents/ runs fine with just checkout + setup-go (the package only imports test/azdcli and test/recording, no built extension is needed). As placed, it runs after the inspector and agents build/publish steps, so a bad selector won't surface until several minutes of extension builds have already run. Moving this step up to right after setup-go would fail fast on selector drift before the expensive build work. Non-blocking.

[glharper]

Addressed in a1f99f3. The selector validation now runs immediately after actions/setup-go, before building azd, the inspector dependency, or the agents extension, so selector drift fails fast. Re-ran the guard locally: it still validates the current 4 Tier 0 / 5 Tier 1 split.

[Copilot]

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated no new comments.