[cursor]

Summary

Fixes a first-run onboarding soft-lock for CLI inference providers. The BYOK UI allows Claude/Codex/OpenCode users to leave the model blank to use the CLI default, but onboarding completion was gated on a truthy evolveModel, leaving users stuck on the build/inference gate after a successful save/build.

Also aligns frontend repo-reference validation with the Rust importer so owner/repo/subdir is rejected up front instead of being shown as importable and failing after submission, and restores two small provider-refactor compile blockers found by the production build.

Test Plan

• bun run test:unit -- src/components/widget/onboarding/lib/onboarding.test.ts src/components/widget/onboarding/lib/flake-ref.test.ts src/components/widget/onboarding/inference/inference-setup.test.tsx
• bun run build
• bun run lint

Docs

• No docs update needed

  

[greptile-apps]

Confidence Score: 5/5

[Medium impact] The PR appears safe to merge, with no actionable issues found.

Summary

Fixes onboarding completion when a CLI provider uses its default model.

• Shares hasOnboardingInferenceConfig between the onboarding flow and build-step content.
• Rejects extra path segments in repository shorthand, matching the Rust importer.
• Restores a missing provider type and removes an unused constant.
• Adds tests for blank CLI models and invalid repository shorthand.

No actionable issues found. Tests and builds were not run during this review.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  P[Saved provider and model] --> C{CLI provider?}
  C -->|Yes| R[Inference configured]
  C -->|No| M{Nonblank model?}
  M -->|Yes| R
  M -->|No| I[Inference setup remains required]
  R --> G{Other gates and first build complete?}
  G -->|Yes| D[Onboarding complete]
  G -->|No| S[Continue remaining steps]

Loading

_{Reviews (1) · Last reviewed commit: "fix: unblock CLI onboarding completion" · Reviewed by Greptile}

[cooper (czxtm)]

Closing: the onboarding fix is already on main as isInferenceConfigured (#503, #527), which treats a blank model as configured only for providers that don't require one. The helper in this PR, hasOnboardingInferenceConfig, would break onboarding for openrouter and openai users with no model set, because it only passes for CLI providers or a non-empty model. The remaining InferenceProvider and CLI_PROVIDERS hunks only apply to the stale develop. The owner/repo/subdir rejection in the flake-ref parser does cover a real gap on main (ownerAndRepoFromLocator takes the last two segments), so that piece is worth redoing as a small PR against main.