You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
test(desktop): stabilize collaboration mode updates during Runtime Host session churn #2545
The Desktop E2E test slash suggestions: project gating, collaboration modes, and Deep Research filtering intermittently fails while switching a Session to Plan mode. The Runtime Host Session identity keeps changing throughout the existing bounded poll, so sessions:setCollaborationMode never converges.
This occurred on a PR that only changes activity-card rendering and its E2E coverage. The other E2E shard and all non-E2E CI lanes passed, which points to a timing-sensitive Runtime Host lifecycle race rather than a deterministic regression in that PR.
The shard result was 19 passed and 1 failed. The sibling E2E shard passed.
Expected behavior
A collaboration-mode update should either:
target a stable Runtime Host Session and complete; or
cross an explicit readiness/reconnect boundary and retry against the canonical replacement Session.
The caller should not exhaust its bounded wait because Session identity continues changing without a stable point.
Investigation scope
Identify why the Session is repeatedly replaced while this test updates collaboration mode.
Determine whether the production client needs a canonical reconnect-and-retry contract or the E2E fixture is missing a real readiness boundary.
Preserve bounded failure behavior; do not hide the race with an unbounded retry or a fixed delay.
Add a deterministic regression at the layer that owns Session replacement and update routing.
Reproduction
npm run e2e --workspace @maka/desktop -- \
--grep "slash suggestions: project gating, collaboration modes, and Deep Research filtering" \
--repeat-each=10
Activity