The Copilot Chat sanity tests in extensions/copilot hit the live
model endpoint, so they can fail for transient upstream reasons such
as empty responses, intermittent rate limiting, or model errors that
do not represent an actual regression.
- Bump mocha retries for sanity tests from 1 to 2 (3 attempts total)
so a single transient failure no longer fails the build.
- Add a shared describeOutcome() helper that dumps the response
stream items and any errorDetails into assertion messages, so the
next failure (after retries are exhausted) carries enough context
to distinguish a flake from a real bug.
- Reuse the helper in the existing agent-mode timeout diagnostics
instead of duplicating the dump logic.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Ulugbek Abdullaev
5b0a8c2b3f