[dependabot]

Bumps blaxel from 0.2.50 to 0.4.17.

Release notes

Sourced from blaxel's releases.

v0.4.1

Fixed

• Job task exceptions now exit non-zero, so Blaxel executions are correctly marked failed instead of falsely reporting success (ENG-4261, #208).

Jobs with retries configured will now retry when their task raises. Tasks that perform non-idempotent work should remain retry-safe.

v0.4.0

⚠️ Behavior change — read before upgrading

Failed sandbox update_* and delete calls now raise SandboxAPIError instead of silently returning a malformed object.

Every generated control-plane API function returns Union[Error, T] | None, so an error status is a return value, not an exception. SandboxInstance.update_metadata, update_ttl, update_lifecycle, update_network and delete passed that response straight into cls(response), which does not type-check. A 403/404/500 therefore produced an instance wrapping an Error.

If you ignored the return value — the common shape — a failed write looked like a success:

• update_ttl returned normally and the TTL was never applied
• delete returned normally and the sandbox stayed alive and kept billing

If you used the return value, you hit AttributeError: 'Error' object has no attribute 'metadata' far from the real cause.

What this means for you

you	impact
Your calls succeed	No change. Same objects, same types, same behavior.
Your calls have been failing silently	You now get exceptions. You were already broken — the update never applied, the delete never happened.
You wrote a workaround checking isinstance(result.sandbox, Error)	This breaks. The call now raises before returning, so your check never runs.

Most notably: delete() on an already-deleted sandbox now raises where it previously returned quietly. If you delete in a finally: or a cleanup path, wrap it:

try:
    await sandbox.delete()
except SandboxAPIError:
    pass  # already gone

Also: delete() on an empty response now raises SandboxAPIError rather than ValueError.

This brings Python in line with the TypeScript SDK, which has always used throwOnError: true on these calls, and with Go, which returns idiomatic (res, err). Python was the only SDK swallowing these errors.

Fixes

• fix(sandbox): propagate control-plane errors from update_*/delete (#206, ENG-4244)
  • all four update_* helpers and delete, in both the async and sync (SyncSandboxInstance) trees
  • delete no longer raises a misleading ValueError("Sandbox X not found") for what may have been a 500
  • VolumeInstance.delete had the same gap and now raises on both Error and empty responses
  • Reported by Albert at EvenUp, who also identified the correct fix: routing through the existing _unwrap_response() helper.

... (truncated)

Commits

• c921c88 feat(image): choose the build region on build() (#281)
• 4f86a75 feat(image): name build sizes build_memory and build_volume (#280)
• bedbc75 fix(image): only send build headers the upload URL was signed for (#279)
• 0a61fd5 feat(image): set build worker memory and volume on build() (#278)
• 9b95667 test: align lifecycle tests with the 5m minimum sandbox TTL (#277)
• d6433ee chore: regenerate SDK clients from latest API definitions (#276)
• 9f7a4dd feat(sandbox): add preview shorthand and expiry-capped token reuse (#264)
• 80de89b feat: add compatible typed API errors (PM-2046) (#263)
• b125dd6 feat(core): settings.integration appends an integration product token to the ...
• be587ad sandbox: fork() takes a lifecycle (#259)
• Additional commits viewable in compare view

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-22T20:45:13.324617Z	3073fd0	New commits
🔒 Security Review	✅ Completed 2026-10-22T20:46:06.160634Z	3073fd0	New commits

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 0a57e8a337

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Keep MCP v2 in the default test environment

This update also downgrades the locked mcp package from 2.0.0 to 1.30.0. Because CI sets UV_FROZEN=1 and make sync installs this lock with all extras, the main test, lint, and type-check jobs now exercise only the MCP v1 implementation; the separate compatibility job also pins v1 versions, while the detailed unit tests are the only coverage of the v2-specific mcp.Client path. A regression in the documented v2 support can therefore pass CI, so retain MCP v2 in the default environment or add an explicit v2 test job before accepting this transitive downgrade.

AGENTS.md reference: AGENTS.md:L19-L19

Useful? React with 👍 / 👎.

[jbeckwith-oai]

The Blaxel update is blocked by its upstream dependency contract: 0.4.13 requires mcp>=1.9.4,<2.0.0, downgrading the default environment from MCP v2 to v1. This breaks the v2 test-server constructors and removes default v2 coverage. I also checked the latest published Blaxel release, 0.4.19: it still declares the same MCP <2 constraint (https://pypi.org/pypi/blaxel/0.4.19/json).

Blaxel is an optional hosted sandbox backend, installed with openai-agents[blaxel]; it is not a core dependency. Retaining the current version keeps the existing backend available without accepting this downgrade. Resolving this requires upstream MCP v2 support or an explicit decision to split the dependency/test environments; suppressing type errors or overriding the upstream constraint would not establish compatibility. No code changes made to this PR.