fix(code-review): add model diagnostics

[alex-alecu]

Summary

• Adds private runtime diagnostics when code reviews fail because a model is unavailable, so support can see the requested model, live model count, and close matches without showing those details publicly.
• Keeps GitHub, GitLab, and public Cloud Agent messages generic while the signed-in dashboard gets a clearer private error.
• Does the live model lookup in the wrapper and gives it a short timeout so a slow lookup cannot block failure reporting.
• Expands isModelNotFoundMessage regex to also match unknown model and invalid model patterns, keeping it in sync with classifyAssistantFailureMessage in safe-failure-projection.ts so those error variants no longer skip diagnostics.

Verification

Manual staging checks:

1. Start a code review with a model that is not available.
2. Confirm the signed-in dashboard shows the missing model, model count, and up to five alternatives.
3. Confirm GitHub or GitLab only shows the generic model-not-found message.
4. Search logs and Sentry by the review id and confirm they include the full runtime model list.

Visual Changes

N/A

Reviewer Notes

• The full model list is private diagnostic data and should not appear in public review output.
• The wrapper drops diagnostics if the live model lookup is slow, so terminal failure reporting keeps moving.
• isModelNotFoundMessage now mirrors the regex pattern in classifyAssistantFailureMessage: /\b(model (?:was )?not found|unknown model|invalid model)\b/i.

Built for alex-alecu by Kilo

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

Executive Summary

The incremental commit (957f429) correctly moves modelNotFoundRuntimeDiagnostics into the WrapperTerminalFailure type, adds a guarded async fetch with a 1 s timeout, and wires the diagnostics through the ingest stream event. All previously flagged issues are resolved; no new issues detected.

Resolved Issues

File	Line	Issue	Status
services/cloud-agent-next/wrapper/src/connection.ts	811	isModelNotFoundMessage regex was narrower than classifyAssistantFailureMessage — unknown model/invalid model variants now covered	✅ Fixed
services/cloud-agent-next/wrapper/src/connection.ts	773	getTerminalFailure was not inspecting the full error object for billing markers	✅ Fixed

Other Observations (not in diff — carried forward, still open)

These were noted in earlier reviews and are not blocking, but remain unaddressed:

File	Note
apps/web/src/app/api/internal/code-review-status/[reviewId]/route.ts (lines 119–198)	Contains a local copy of ModelNotFoundRuntimeDiagnostics type and parseModelNotFoundRuntimeDiagnostics with hardcoded constants (MAX_MODEL_ID_LENGTH = 512, MAX_SUGGESTIONS = 5) that duplicate services/cloud-agent-next/src/shared/runtime-model-diagnostics.ts. If the shared module's limits change, the web app's validation silently diverges.
services/cloud-agent-next/src/websocket/ingest.ts:70 and services/cloud-agent-next/src/session/wrapper-supervisor.ts:56	MODEL_NOT_FOUND_SAFE_ERROR_MESSAGE = 'Assistant request failed: model not found' is copy-pasted rather than imported from a single source.

Files Reviewed (24 files)

• services/cloud-agent-next/wrapper/src/connection.ts — previously 2 issues, now resolved
• services/cloud-agent-next/wrapper/src/main.ts — no issues
• services/cloud-agent-next/wrapper/src/model-diagnostics.ts — no issues
• services/cloud-agent-next/wrapper/src/kilo-api.ts — no issues
• services/cloud-agent-next/src/shared/protocol.ts — no issues
• services/cloud-agent-next/src/shared/runtime-model-diagnostics.ts — no issues
• services/cloud-agent-next/src/shared/index.ts — no issues
• services/cloud-agent-next/src/session/wrapper-supervisor.ts — no issues
• services/cloud-agent-next/src/session/message-settlement-outbox.ts — no issues
• services/cloud-agent-next/src/session/session-message-state.ts — no issues
• services/cloud-agent-next/src/websocket/ingest.ts — no issues
• services/cloud-agent-next/src/callbacks/types.ts — no issues
• services/cloud-agent-next/test/unit/wrapper/model-diagnostics.test.ts — no issues
• services/cloud-agent-next/test/unit/wrapper/reconnection.test.ts — no issues
• services/cloud-agent-next/test/unit/wrapper/server.test.ts — no issues
• services/cloud-agent-next/test/unit/wrapper/snapshot.test.ts — no issues
• services/cloud-agent-next/test/unit/wrapper/kilo-api.test.ts — no issues
• services/cloud-agent-next/wrapper/package.json — no issues
• apps/web/src/app/api/internal/code-review-status/[reviewId]/route.ts — duplicate parser noted above
• apps/web/src/app/api/internal/code-review-status/[reviewId]/route.test.ts — no issues
• services/cloud-agent-next/src/session/message-settlement-outbox.test.ts — no issues
• services/cloud-agent-next/src/session/wrapper-supervisor.test.ts — no issues
• services/cloud-agent-next/src/websocket/ingest.test.ts — no issues
• pnpm-lock.yaml — skipped (generated)

---

_{Reviewed by claude-4.6-sonnet-20260217 · 3,649,342 tokens}

_{Review guidance: REVIEW.md from base branch main}

[alex-alecu]

@eshurakov how do you feel about adding a new dependency? When a model is not found, I do a fuzzy search on the available runtime (CLI) models against the one set in web and return just a few.