feat: add journaled Cline CLI integration and catalog sync

[lidge-jun]

Summary

• Add Cline CLI/shared SDK provider-store detection, routed catalog export, existing integration verbs, and dashboard entries. The integration uses the source-backed version-1 providers.json/models.json schema; legacy VS Code extension storage is not migrated.
• Manage only the OpenCodex entries and preserve the default provider. One existing journal operation snapshots both original files, with per-file atomic replacement, compensation and durable interrupted-operation recovery. Stop Cline before mutations and restart afterward; explicit sync refreshes owned pairs, while unattended refresh excludes Cline.
• Explain the two-document download and Undo contract in the existing dashboard and guides. Reuse the existing Cline mark. Addresses [Feature]: add Cline as a supported client integration #4214; feat(clients): add native Command Code integration and catalog sync #3833 is a separate Command Code reference, with no dependency or carried implementation.

Verification

• Local product tests (all sizes), typecheck, builds and installation: NOT RUN, explicitly prohibited by the user. Push used git push --no-verify; no global hook settings were changed.
• git diff --check: passed (text-only check).
• Inherited independent source/design and implementation reviews completed. Recovery findings were corrected and source-reviewed with no remaining blockers in that scope.
• Added source-backed config/path fixtures and paired writer tests for exact Undo/absence, protected edits, model retirement, second-file/bookkeeping failures, pending recovery, malformed authority and strict-read uncertainty. Added dashboard bundle-copy and client-map coverage.
• Final cumulative head: c08d792c5f15a5ec55c8bd6384ac9ee35ebc02f8. Hosted Cross-platform CI is queued/pending; no test pass is claimed yet. No intermediate tip is claimed tested.
• Actual source components rendered with synthetic status/history in Chrome. Existing dependencies were reused; only the isolated preview entry was bundled (19–20 ms), with no local product suite, typecheck, installation or full build. GUI source tree for the final export capture: a070b75cabe1974e59d407c595709d1ffb3c4a58.

Checklist

• Scope stays focused and avoids unrelated cleanup.
• Docs or release notes were updated when needed.
• Security-sensitive changes were reviewed for secrets, auth, and unsafe defaults.

Delivery: one cohesive ordinary PR targeting dev; no native stack and no merge/auto-merge requested. The coordinating task owns integration and issue closure. No original PR was carried; no contributor attribution is invented.

Maintainer integration decision

Owner-authorized dev integration at c08d792c5f15a5ec55c8bd6384ac9ee35ebc02f8 with all local suites prohibited and final-tip hosted CI tracked after delivery. Independent parent-side reviewer Linnaeus accounted for all 67 original paths, reviewed all 47 source/test/layout files plus primary source/owner docs, and returned scoped source/security PASS. Recovery refuses unreadable or inconsistent journals/ownership; foreign entries and default selection remain intact; writes are explicit-only and Cline must be stopped. Pinned upstream provider-store schema was independently fetched and byte-matched.

The final rebase/product-test normalized patches are identical; Linnaeus separately reviewed the final five-file UI/screenshot delta and returned PASS with no new blockers. Coordinator inspected the Cline-only label guard and actual refreshed synthetic export screenshot. Screenshots are component observations with fictional data, not actual Cline runtime/Windows or full-system proof. Current code uses the CLI/shared SDK store; legacy extension storage is not migrated.

Final-tip hosted run https://github.com/lidge-jun/opencodex/actions/runs/34675421730 is pending, not claimed passed. Local suites/build/typecheck/install NOT RUN; the documented small source-preview bundling was only for synthetic component capture. No real user Cline configuration or live OpenCodex service was changed. Runtime/Windows/restart acceptance and any trailing failure remain explicit verification obligations. No outstanding review threads or maintainer change requests at intake; no original implementation author was invented. This is a maintainer integration decision, not self-approval or CI success.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

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.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-12T05:26:19.196492Z	c08d792	Draft marked ready

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.