Update JaCoCo coverage badge

[github-actions]

Automated JaCoCo coverage badge update from CI.

[edburns]

@copilot look at the test counts on these three runs corresponding to these three PRs:

1. 3 weeks ago. PR #127 https://github.com/github/copilot-sdk-java/actions/runs/25004480422
   • 817 tests.
2. 4 days ago. PR #189 https://github.com/github/copilot-sdk-java/actions/runs/25884708743
   • 888 tests
3. 49 minutes ago. PR #212 https://github.com/github/copilot-sdk-java/actions/runs/26063266618
   • 998 tests.

These test numbers are monotonically increasing. Yet the coverage percentage is dropping.

Why?

Is the coverage number accurate?

Are we missing some things in the way we track coverage?

[Copilot]

@copilot look at the test counts on these three runs corresponding to these three PRs:

1. 3 weeks ago. PR Fix jbang-example.java shebang line and dependency version #127 https://github.com/github/copilot-sdk-java/actions/runs/25004480422 ...

I checked the three workflow runs and the current config. The badge number is accurate for what we currently measure: JaCoCo instruction coverage from the Java SDK Tests job, generated from target/site/jacoco-coverage/jacoco.csv after mvn verify.

Why it dropped while tests went up: the denominator grew faster than the tests improved executed coverage. PR #212 in particular pulled in a large reference sync with many new/generated classes under com.github.copilot.sdk.generated and com.github.copilot.sdk.generated.rpc. The current report is 78.3% overall, but generated.rpc is only 64.6%, and several newly added generated types are at 0%.

So I don’t think the badge script is wrong. I do think our tracking has a couple of blind spots:

• the badge is instruction coverage only (not line/branch/method/class)
• it mixes generated code with handwritten SDK code
• it only uses the Java SDK Tests JaCoCo exec; the separate smoke-test jobs are not merged into that number

If we want the badge to better reflect SDK logic quality, the follow-up change should be to exclude generated packages from the badge or publish separate coverage numbers for generated vs handwritten code.

[edburns]

@copilot wrote:

If we want the badge to better reflect SDK logic quality, the follow-up change should be to exclude generated packages from the badge or publish separate coverage numbers for generated vs handwritten code.

Please work into this PR the change to do the latter. I want separate coverage numbers for generated vs. handwritten code, and badges for both.

[Copilot]

@copilot wrote:

If we want the badge to better reflect SDK logic quality, the follow-up change should be to exclude generated packages from the badge or publish separate coverage numbers for gen...

Addressed in e7da399. The PR now generates separate JaCoCo badges for handwritten and generated code, updates the README to show both, and adds the same handwritten/generated instruction-coverage split to the CI test summary.