feat(improve-prompt): Update BFF route for evaluations

[kaapi-agent-bot]

Issue

Closes #273

Summary

• Repoint the improve-prompt BFF route to POST /api/v2/evaluations/{evaluation_id}/improve-prompt instead of v1.
• Maintain existing callback_url resolution and /api/webhooks/prompt-improvement behavior.
• Type webhook callback payload as PromptRecommendationJobPublic with recommendation_type: "prompt" and reject other types.
• Display a friendly toast for 422 not_a_judge_run response from the backend.
• Disable "Iterate on Prompt" button for v1 runs with a tooltip explaining the reason.

Checklist

Before submitting a pull request, please ensure that you mark these task.

• Ran npm run dev and npm run build in the repository root and test.
• If you've fixed a bug or added code that is tested

Original PR description

Summary

• Repoint the improve-prompt BFF route at POST /api/v2/evaluations/{evaluation_id}/improve-prompt instead of v1.
• Keep the existing callback_url resolution and /api/webhooks/prompt-improvement receiver behavior unchanged (body shape is the same).
• Type the webhook callback payload as PromptRecommendationJobPublic with a recommendation_type: "prompt" discriminant (ready to widen to a union of recommendation types later), and reject payloads where it isn't "prompt".
• Surface a friendly toast for a 422 not_a_judge_run response from the backend.
• Disable the "Iterate on Prompt" button client-side for v1 (cosine-only) runs — judged runs are detected via the existing isNewScoreObjectV2 check (v2 score objects carry summary_scores/traces, v1 only has cosine_similarity) — with a tooltip explaining why.

Assumptions

• The v2 endpoint accepts the same request body shape as v1 (callback_url plus any pass-through fields), per the issue's "body shape unchanged" note. Not verified against the live backend since this run has no network access.
• The 422 error response for non-judged runs follows the existing pattern used elsewhere in this route ({ error: "not_a_judge_run" }), matching how other sync-validation codes (e.g. evaluation_not_found) are already surfaced by apiFetch/extractErrorMessage.
• The webhook callback's data object is PromptRecommendationJobPublic with recommendation_type: "prompt" as the only current variant (per the issue's note "widening to a union later"); only job_id, status, recommendation_type, config_version, and error_message are read, mirroring the previous PromptImprovementJobPublic shape.
• Client-side "judged run" detection reuses isNewScoreObjectV2 (already used elsewhere in this page to distinguish v2/judge score shapes from legacy cosine-only v1 shapes) rather than adding a new backend field, since no explicit is_judge_run/judge_type field exists in EvalJob today.

Closes #273

🤖 Generated with Claude Code

[coderabbitai]

Important

Review skipped

Auto reviews are limited based on label configuration.

🏷️ Required labels (at least one) (1)

• ready-for-review

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Advanced

Run ID: 7f8fc5fd-f3ed-49ec-ac9f-7d8aa4f1db00

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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