[WRONG BRANCH] fix(cursor): gate synthetic Ultra (1m) catalog rows on discovered Max Mode capability

[luvs01]

Motivation

• Cursor discovery collects a maxModeModels capability but the catalog filter admitted synthetic Ultra rows (e.g. kimi-k3-1m) whenever the base model was present, falsely advertising 1M/Max Mode when the account lacked Max Mode.
• The change intends to make the authoritative catalog honest about Max Mode availability so clients are not shown unsupported Ultra picker rows that would later produce rejected requests.

Description

• Require the discovered Max Mode capability when admitting synthetic Ultra rows by extending filterCursorConfiguredModelsByLiveDiscovery with a maxModeLiveIds parameter and checking it for Ultra entries in src/adapters/cursor/discovery.ts.
• Propagate liveResult.maxModeModels from discovery into the catalog filter call in src/codex/catalog/provider-fetch.ts so the capability signal is consulted when discovery succeeds.
• Add regression tests in tests/cursor-ultra-mode.test.ts that assert the synthetic kimi-k3-1m row is included only when Max Mode is discovered and hidden when Max Mode is absent.
• Keep the existing degraded/static fallback behavior unchanged: discovery failures still fall back to the configured static seed per existing policy.

Testing

• Ran bun test tests/cursor-ultra-mode.test.ts tests/cursor-discovery.test.ts tests/cursor-uncallable-quarantine.test.ts and observed all tests in those files passed (26 passed, 0 failed).
• Ran bun run typecheck and the repository typecheck completed successfully.
• Ran the full suite with bun run test and validated the change; the overall run encountered two unrelated failures in tests/lab-evidence-sanitization.test.ts that are not caused by this patch and were observed during the full-suite run.
• Verified git diff --check and focused repository checks before committing the three modified files: src/adapters/cursor/discovery.ts, src/codex/catalog/provider-fetch.ts, and tests/cursor-ultra-mode.test.ts.

---

Codex Task

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: 88d375e3-5f63-4c8a-9c98-776b8e3f2428

---

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.}

[github-actions]

✅ Deterministic PR hygiene checks passed.

[github-actions]

⏳ DRAFT

• wrong target branch (main); retarget to dev.

What to do

• Retarget this PR to dev — all contributions go to dev.

Its title has been prefixed with [WRONG BRANCH].
This pull request was already a draft. Its draft status will be preserved after every issue above is resolved.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b6fbf39911

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Preserve capability filtering when every configured row is rejected

When a Cursor provider explicitly configures only kimi-k3-1m and successful discovery returns a Kimi base model without any maxModeModels, this filter correctly produces an empty list, but the immediately following available.length > 0 ? available : configured restores and caches the unsupported synthetic row. The catalog therefore still advertises the 1M selector to accounts lacking the discovered capability; a successful nonempty discovery should remain authoritative even when no configured rows survive, rather than falling back to configured.

Useful? React with 👍 / 👎.