Add retry to exec()

[mendral-app]

Summary

• Wrap SandboxProcess.exec() POST call with retry_on_transient_reset_async, matching the pattern used by get(), list(), and logs() in the same file
• Prevents httpx.ReadError from stale connections propagating unhandled to callers

Context

Insight: https://app.mendral.com/insights/01M07AAN6MCM0S1MWHKDGTH1XZ

exec() was the only sandbox method missing the transient retry wrapper, causing unhandled ReadError exceptions when the httpx client held a stale TCP connection.

---

Note

Created by Mendral. Tag @mendral-app with feedback or questions.

• View implementation session
• Addresses this insight

---

Note

Low Risk
Small, localized resilience change aligned with existing sandbox HTTP patterns; no auth or data-model changes.

Overview
SandboxProcess.exec() now wraps its POST /process HTTP call with retry_on_transient_reset_async, matching get(), list(), and logs() in the same module.

That closes a gap where stale pooled connections could raise unhandled httpx.ReadError (and related transient failures) on exec while other process APIs already retried and reset the client.

^{Reviewed by Cursor Bugbot for commit dded5d3. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

Want fixes drafted automatically? Bugbot Autofix can create code changes for findings. A team admin can enable Autofix in the Cursor dashboard.

^{Reviewed by Cursor Bugbot for commit 5ee526d. Configure here.}

[mendral-app]

Thanks for the catch — you're right that POST /process is non-idempotent and retrying could spawn duplicate processes if the server accepts the request but the connection drops before the response arrives.

I've reverted the retry wrapper in dded5d3, restoring the original non-retried path. This is consistent with:

• The is_transient_reset_error docstring scoping retries to idempotent calls
• The sandbox_read_retries setting name
• The existing test_process_exec_is_not_retried_on_transport_reset assertion

The original insight about stale connection errors propagating from exec() remains valid, but the correct mitigation would need to be at a different layer (e.g., connection pool health checks or client keepalive tuning) rather than blindly retrying a mutating operation.