[ERR_HTTP2_INVALID_SESSION]: The session has been destroyed — causes repeated transient retries

[sriadd]

Describe the bug

The CLI surfaces Request failed due to a transient API error. Retrying... followed by:

✗ Error [ERR_HTTP2_INVALID_SESSION]: The session has been destroyed

This happens frequently (multiple times per session, often mid-turn on long reasoning responses). I would then be unable to issue a /restart command, and have to quit the cli and relaunch is with resume to continue. And then would encounter the issue again frequently

Affected version

1.0.47

Steps to reproduce the behavior

Steps to reproduce

1. Launch copilot in a normal repo.
2. Run any non-trivial prompt that produces a long reasoning turn (e.g. multi-file edit, /research,
   planning).
3. Within the turn, the HTTP/2 session error fires; CLI retries.

The error is non-deterministic but reproduces within ~5–15 minutes of active use.

What I've already tried

• /restart — temporarily clears, error returns later.
• /update — already on latest (1.0.47).
• Switched models via /model — happens on multiple models.
• Forced IPv4 to rule out a broken IPv6 path: NODE_OPTIONS="--dns-result-order=ipv4first" copilot

Verified the flag is applied. Error still occurs.

Expected behavior

No response

Additional context

OS macOS 15.7.7
arm64 (Apple Silicon)
Node version v22.22.0
terminal: cmux
shell: zsh