fix(inference): use model-aware token field in rebuild preflight

[laitingsheng]

Summary

The atomic DCode rebuild gate probes the sandbox's recorded OpenAI-compatible route before making changes. Its Chat Completions payload hard-coded max_tokens, so rebuilds configured with GPT-5 or o-series models could fail with HTTP 400 before rebuild processing began.

This PR makes that rebuild-only probe use the shared model-aware token-field resolver. It does not change the managed gateway's ongoing route-health validation.

Scope

preflightRebuildInferenceRoute is called only by the DCode rebuild preflight. This PR intentionally does not change the managed gateway's ongoing route-health validation or claim broader runtime behavior.

Changes

• Resolve the Chat Completions reply-budget field with resolveMaxTokensField(input.model).
• Send max_completion_tokens for GPT-5 and o1/o3/o4 model families.
• Preserve max_tokens for models that support the legacy field.
• Cover GPT-5, o-series, and legacy-model behavior at the rebuild-preflight command boundary.

Type of Change

• Code change (feature, bug fix, or refactor)
• Code change with doc updates
• Doc only (prose changes, no code sample modifications)
• Doc only (includes code sample changes)

Quality Gates

• Tests added or updated for changed behavior
• Existing tests cover changed behavior - justification:
• Tests not applicable - justification:
• Docs updated for user-facing behavior changes
• Docs not applicable - justification: internal rebuild inference preflight behavior; no user-facing contract changed.
• Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging)
• Sensitive-path review completed or maintainer-approved waiver recorded - reviewer/approval link/justification:
• Non-success, skipped, or missing CI check accepted by maintainer - check name, approval link, and follow-up issue:

Verification

• PR description includes Signed-off-by: lines and every commit is signed
• npm run check:diff
• npm run build:cli
• npm run typecheck:cli
• npx vitest run --project cli src/lib/actions/sandbox/rebuild-inference-preflight.test.ts - 7 passed
• Quality Gates section completed with required justifications or waivers
• No secrets, API keys, or credentials committed

Signed-off-by: Tinson Lai tinsonl@nvidia.com
Signed-off-by: Prekshi Vyas prekshiv@nvidia.com

Summary by CodeRabbit

• Bug Fixes
  • Improved inference preflight checks for newer GPT-5 and o-series models by using the correct completion token parameter.
  • Preserved compatibility with legacy models by continuing to use their supported token limit parameter.
  • Added coverage to verify request payloads use the appropriate field for each model type.

[coderabbitai]

📝 Walkthrough

Walkthrough

The inference preflight probe now selects the max-token field based on the model. Tests verify GPT-5 and o-series models use max_completion_tokens, while legacy models use max_tokens.

Changes

Inference preflight token selection

Layer / File(s)	Summary
Model-aware probe payload and coverage src/lib/actions/sandbox/rebuild-inference-preflight.ts, src/lib/actions/sandbox/rebuild-inference-preflight.test.ts	The probe payload resolves the max-token field from the model, with tests covering GPT-5, o-series, and legacy chat-completions requests.

Estimated code review effort: 2 (Simple) | ~10 minutes

Suggested labels: provider: nvidia

Suggested reviewers: cv, ericksoa, cjagwani

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly summarizes the main change: using a model-aware token field in the rebuild preflight probe.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/preflight-gpt5-reply-budget

---

_{Comment @coderabbitai help to get the list of available commands.}

[github-code-quality]

Code Coverage Overview

Languages: TypeScript

TypeScript / code-coverage/plugin

The overall coverage remains at 96%, unchanged from the main branch.

TypeScript / code-coverage/cli

The overall coverage in the fix/preflight-gpt5-r... branch remains at 79%, unchanged from the main branch.

Show a code coverage summary of the most impacted files.

File	main 20130dd	fix/preflight-gpt5-r... 7b7640f	+/-
src/lib/agent/dashboard-ui.ts	89%	83%	-6%
src/lib/state/config-io.ts	94%	91%	-3%
src/lib/agent/m...fest-readers.ts	97%	95%	-2%
src/lib/state/gateway.ts	93%	91%	-2%
src/lib/messagi...annel-config.ts	99%	98%	-1%
src/lib/state/sandbox.ts	85%	85%	0%
src/lib/messagi.../persistence.ts	86%	89%	+3%
src/lib/messagi...n-validation.ts	96%	100%	+4%
src/lib/sandbox...rce-identity.ts	89%	94%	+5%
src/lib/adapter...shell/client.ts	83%	88%	+5%

---

_{Updated July 14, 2026 17:28 UTC
Code Coverage is in Public Preview. Learn more and provide us with your feedback.}

[coderabbitai]

🧹 Nitpick comments (1)

src/lib/actions/sandbox/rebuild-inference-preflight.test.ts (1)

67-80: 🎯 Functional Correctness | 🔵 Trivial | ⚡ Quick win

Cover the o-series branch as well.

The resolver also maps o1/o3/o4 models to max_completion_tokens, but these action-level tests cover only GPT-5 and the fallback branch. Add an o-series case to protect the full model-selection contract.

Suggested test

+  it("sends max_completion_tokens for an o-series model on the chat completions route (`#6850`)", () => {
+    const command = buildRebuildInferenceProbeCommand({ ...input, model: "o3-mini" });
+
+    expect(command).toContain('"max_completion_tokens":8');
+    expect(command).not.toContain('"max_tokens"');
+  });

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/lib/actions/sandbox/rebuild-inference-preflight.test.ts` around lines 67
- 80, Add an action-level test alongside the existing GPT-5 and non-GPT-5 cases
in the resolver test suite, using an o-series model such as o1, o3, or o4.
Verify the generated chat-completions command contains max_completion_tokens
with the expected value and excludes max_tokens, covering the resolver’s
o-series branch.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Nitpick comments:
In `@src/lib/actions/sandbox/rebuild-inference-preflight.test.ts`:
- Around line 67-80: Add an action-level test alongside the existing GPT-5 and
non-GPT-5 cases in the resolver test suite, using an o-series model such as o1,
o3, or o4. Verify the generated chat-completions command contains
max_completion_tokens with the expected value and excludes max_tokens, covering
the resolver’s o-series branch.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 335b4448-bbdd-4793-bbe8-97c1ad0b04e5

📥 Commits

Reviewing files that changed from the base of the PR and between 33d55b0 and fa5dcdc.

📒 Files selected for processing (2)

• src/lib/actions/sandbox/rebuild-inference-preflight.test.ts
• src/lib/actions/sandbox/rebuild-inference-preflight.ts

[github-actions]

PR Review Advisor — Informational

Advisor assessment: Informational / high confidence
Next action: No advisor follow-up needed.
Findings: 0 blockers · 0 warnings · 0 suggestions
Status: No actionable findings remain in the canonical review ledger.

Model lanes

• GPT-5.6 Terra (primary): Completed · high confidence · 0 blockers · 0 warnings · 0 suggestions
• Nemotron 3 Ultra (second opinion): Completed · high confidence · 0 blockers · 0 warnings · 0 suggestions
• Model comparison: normalized findings match; normalized E2E selections differ; severity counts match.

Nemotron output stays in workflow artifacts and does not change the assessment above.

E2E guidance

Advisory only. E2E / PR Gate selects and runs jobs independently.

Recommended E2E: inference-routing, network-policy, onboard-repair, onboard-resume, state-backup-restore, upgrade-stale-sandbox

1 optional E2E recommendation

• sandbox-rebuild

Workflow run details

This automated review informs maintainers. Warnings and suggestions do not require a response. A maintainer decides whether to merge.