## Description

Adds a new Copilot skill (`evaluate-pr-tests`) that evaluates the
quality, coverage, and appropriateness of tests added in PRs.

## Why This Is Useful

**Problem:** When reviewing PRs, evaluating test quality is one of the
most time-consuming and inconsistent parts of the process. Common issues
slip through:

- Tests that only verify a page loads but never exercise the actual fix
- UI tests for behavior that could be caught by a unit test (which runs
in seconds vs minutes)
- Missing edge cases (null inputs, boundary values, repeated actions)
that lead to regressions later
- Convention violations (wrong base class, missing attributes, obsolete
APIs) that cause CI failures
- Flaky patterns (arbitrary delays, missing `WaitForElement`) that waste
CI time

These issues are tedious to catch manually and easy to miss — especially
across 4 platforms, 4 test types (unit, XAML, device, UI), and hundreds
of existing test patterns.

**Solution:** This skill automates the mechanical checks and guides the
agent through deeper analysis:

1. **Automated (script):** File categorization, convention compliance,
anti-pattern detection, AutomationId consistency, existing similar test
search, platform scope analysis
2. **Agent-guided (SKILL.md):** Fix coverage analysis, edge case gap
identification, test type recommendation (prefer unit > XAML > device >
UI), assertion quality review

**Real-world example:** Running against PR dotnet#34324 (Android Shell TabBar
fix), the skill:
- ✅ Correctly identified 1 fix file + 1 UI test + 1 HostApp page + 1
snapshot
- ⚠️ Flagged an inline `#if` platform directive (real convention
violation)
- ℹ️ Noted 3 HostApp AutomationIds not exercised by the test
(informational)
- ✅ Found existing Shell tests for cross-reference
- ✅ Identified this as Android-only based on fix file paths

**Quality bar:** This skill went through 3 rounds of multi-model review
(10 total reviews across Opus, GPT-5.1, Gemini, Sonnet 4.5) to catch
regex bugs, false positive/negative edge cases, and documentation
mismatches.

## Evaluation Criteria (9 checks)

| # | Criterion | What It Checks |
|---|-----------|---------------|
| 1 | **Fix Coverage** | Does the test exercise the actual code paths
changed by the fix? |
| 2 | **Edge Cases & Gaps** | Boundary conditions, null handling,
async/timing, repeated actions |
| 3 | **Test Type Appropriateness** | Could a lighter test type work?
(unit > XAML > device > UI) |
| 4 | **Convention Compliance** | Naming, attributes, base classes,
obsolete APIs |
| 5 | **Flakiness Risk** | Delays, missing WaitForElement, screenshot
hygiene |
| 6 | **Duplicate Coverage** | Does a similar test already exist? |
| 7 | **Platform Scope** | Does test coverage match platforms affected
by the fix? |
| 8 | **Assertion Quality** | Are assertions specific enough to catch
regressions? |
| 9 | **Fix-Test Alignment** | Do test and fix target the same code
paths? |

## Components

- **`SKILL.md`** — Evaluation criteria with decision trees, examples,
and structured report template
- **`Gather-TestContext.ps1`** — Automated convention checks,
anti-pattern detection, AutomationId consistency, platform scope
analysis
- **`copilot-instructions.md`** — Skill registration with trigger
phrases

## Usage

```bash
# Auto-detect PR context
pwsh .github/skills/evaluate-pr-tests/scripts/Gather-TestContext.ps1

# Or ask the agent
# "evaluate the tests in this PR"
# "are these tests good enough?"
# "review test quality for PR #XXXXX"
```

- Fixes dotnet#34139

---------

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: Jakub Florkowski <42434498+kubaflo@users.noreply.github.com>