fix(router): increase inference validation token budget

[geelen]

Summary

Increase the inference validation probe token budget from 1 to 32 so OpenAI-compatible backends that reject extremely small output budgets can still pass verification.

Related Issue

N/A

Changes

• Increased the validation probe token budget from 1 to 32 for chat completions, completions, Anthropic messages, and responses probes
• Updated the router-side validation test to expect the new probe budget
• Updated the server-side inference verification test to match the new probe request shape

Testing

• mise run pre-commit passes
• Unit tests added/updated
• E2E tests added/updated (if applicable)

Checklist

• Follows Conventional Commits
• Commits are signed off (DCO)
• Architecture docs updated (if applicable)

[github-actions]

All contributors have signed the DCO ✍️ ✅
_{Posted by the DCO Assistant Lite bot.}

[github-actions]

Thank you for your interest in contributing to OpenShell, @geelen.

This project uses a vouch system for first-time contributors. Before submitting a pull request, you need to be vouched by a maintainer.

To get vouched:

1. Open a Vouch Request discussion.
2. Describe what you want to change and why.
3. Write in your own words — do not have an AI generate the request.
4. A maintainer will comment /vouch if approved.
5. Once vouched, open a new PR (preferred) or reopen this one after a few minutes.

See CONTRIBUTING.md for details.

[geelen]

I have read the DCO document and I hereby sign the DCO.

[geelen]

FYI I have now tested this against the particular endpoint and it does indeed pass validation automatically. Also the value of 32 was just plucked out of thin air, but seemed like a safe default (my endpoint returned 11 tokens in response).

[pimlock]

FYI I have now tested this against the particular endpoint and it does indeed pass validation automatically. Also the value of 32 was just plucked out of thin air, but seemed like a safe default (my endpoint returned 11 tokens in response).

I think 32-ish makes sense and shouldn't impact the time it takes for response to come back too much. Flakiness/potential timeouts, etc. was a reason to include the --no-verify flag, so the check is not a blocker.

I just checked how the openclaw does verification and they also use 1 for max_tokens: https://github.com/openclaw/openclaw/blob/757c2cc2deb9a1157a0b5685eaff33bd4bb70485/src/commands/onboard-custom.ts#L269

---

Out of curiosity - what's the validation on the inference-api side? I'm assuming this is some kind of default that litellm is enforcing?