[Prateep07]

Summary
Adds a detailed health://checks resource with overall status, per-check results and latency, uptime, SDK version, timestamp, and transform telemetry. Checks run concurrently with configurable timeouts. The existing health://status response shape remains compatible.

closes #71

Verification

• Health-focused tests: 14 passed
• Full suite: 832 passed, 3 environment-related failures (missing bdist_wheel; regex worker permission errors)

[ROHIT-JR]

Thanks, the concurrent checks, per-check timeouts and the unchanged health://status shape are good. Two things need to change before merge.

Changes needed

1. A hung sync check blocks shutdown. Sync checks now run through asyncio.to_thread. I tried a check that does time.sleep(3) with timeout=0.1. The payload correctly reports "Timed out after 0.1s", but asyncio.run(...) still took 3.0s to finish because it waits for the worker thread. A check that stays hung will also use up the default thread pool as it is polled. Running sync checks in a thread is also a behaviour change for checks that need the main thread (for example a thread-bound DB connection), and the README does not mention it. Please either keep sync checks on the loop (documenting that only async checks are timed out) or run them in a dedicated bounded executor and document the limit.
2. Timeouts are found by string match. The aggregate uses message.startswith("Timed out") to tell a timeout from a false result. Use a distinct per-check status (for example timeout) so the message text can change safely.

Smaller points

• ServerHealthPayload declares transformsPipeline, sandboxWorkerPool and spillover, and nothing fills them in. Remove them or populate them.
• transforms[].details is always {}, so the "transform telemetry" is only a name list for now. Say so in the README or add real details.
• importlib.metadata.version runs on every request, and the fallback is server_config.version (the app's version, not the SDK's). Compute it once and fall back to the SDK's __version__.
• error checks return str(exc) to any client that can read the resource. This is already true for health://status, but consider a generic message here.
• assert elapsed < 0.3 in the timeout test may be flaky on a loaded CI runner.

[ROHIT-JR]

Thanks, both blockers are fixed. Approving.

I re-ran my hung-check script (time.sleep(3), timeout=0.1) on the new head. asyncio.run now returns in 0.1s (it was 3.0s), the check reports status timeout, the other checks are unaffected, and tests/test_health_async_checks.py passes (7). The distinct timeout status, the generic error message on health://checks, the dead payload fields removed, the SDK version resolved once, and the README note about thread-bound resources all look good. CI is green on 3.10 to 3.12.

Two small, non-blocking notes:

• The pool threads are non-daemon, so a check that never returns still delays process exit until it finishes (my script took 3.7s in total). That is much better than before and is a reasonable trade-off, but it is worth one line in the README.
• The __version__ fallback is hard-coded to "0.3.2" and will drift. Reading it from pyproject.toml, or reporting "unknown", would be safer.