feat(runtime): establish native event and transport foundations

[lidge-jun]

Summary

• Add a versioned, trace-first native event/outcome contract with bounded, redacted Codex App and Pi projections. Preserve existing final-response selection when an explicit native outcome is absent.
• Keep authoritative empty/absent finals separate from previews and interrupted-message salvage. Preserve messaging final delivery, ACK and queue-notice behavior, including native body-delivery confirmation.
• Add validated, reversible Cursor/Grok/Claude transport settings and isolated native session namespaces. Existing installations stay in print compatibility mode. Unsupported native main/worker adapters fail before fallback, resume, bootstrap or process work.
• Preserve captured session ownership across compaction, lifecycle persistence, steering and late acquisition failures. Add status implementation flags without replacing cached binary/authentication evidence.

Compatibility and boundaries

Codex App and Pi remain usable through their existing native runtimes. This PR does not activate Cursor/Grok ACP, Claude SDK, or the new Activity presentation default. Those are subsequent runnable layers. Native main and worker support are separate capabilities; unavailable capabilities are not advertised as implemented.

This is the first layer of the planned native-runtime/Activity stack, based on dev. Subsequent layers cover Cursor ACP, Grok ACP, Claude SDK/worker ownership, journal/web presentation, and TUI/desktop. Native stack membership will be registered and verified when the next dependent PR exists. No merge or release is requested by this change.

Verification

• Foundation verification at 7f52d3f10: 329 tests passed across 29 explicitly selected runtime/settings/session regression files; no full repository-wide local suite was run.
• Repair verification at 837bc868b: 169 tests passed across nine focused files. Real 80×24 and 40×24 terminal captures confirm the thinking-to-answer output transition, intact Korean text, single flush, and cleanup.
• Server and frontend typechecks, atomic server build, dist asset verification, frontend production build, architecture count verification (464 entries), and diff checks passed.
• Real compiled-server HTTP/SSE checks covered 13 settings/status/input cases and an unsupported-native request: one actionable error terminal, no print/tool/output fallback, request settlement, then idle state. Test server and resources were torn down.
• Separate independent reviews covered runtime/session code, configuration/DTO code, and the final evidence set.
• Follow-up verification at 157e7a5605bcbfe2ffeafabd6488603ca381bea0: 204 tests passed across 11 explicitly selected files, with server/frontend typechecks, builds, dist assets, architecture counts and diff checks passing. The production tree is unchanged from the recorded PTY capture at 837bc868b.
• Exact-head Tests run and Postinstall Platform Checks both passed at 157e7a5605bcbfe2ffeafabd6488603ca381bea0. Node tests: 8,883 total, 8,837 passed, zero failed, 46 skipped. The clean dev baseline at 348e95f948c4dfa97d4d51ff26711da9dbc838f3 also had zero failures (8,507 passed, 46 skipped); no new failing cases. Windows unit tests, CLI/API smoke checks, and Linux/macOS/Windows-native/WSL installation lanes passed.
• Existing Vite theme-boot and chunk-size warnings remain. CodeRabbit reports review skipped for the base branch; this is not counted as an independent review.

The first CI pass exposed stale Slack mock and source-spelling test contracts. Follow-up verification replaces the affected checks with behavior tests, isolates the TUI fixtures, and fixes a real line-mode thinking-to-answer transition: the first answer chunk now starts its stream sink without resetting the turn clock. Later test-only repairs replace a Mermaid source-window oracle with actual VirtualScroll snapshots and await real delivery promises instead of a fixed number of event-loop yields. Controlled mutations prove that the replay/trace, Mermaid promotion, and delivery-completion oracles catch regressions; every mutation was restored. Final focused, PTY, and exact-head CI evidence is recorded above.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

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: defaults

Review profile: CHILL

Plan: Team

Run ID: db0cf1d7-4b82-4804-a44c-6f4236d2425d

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.}

[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 2026-09-05T13:29:44.763654Z	b717b24	PR opened

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.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b717b249c2

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]

Remove settled preview rows before applying native finals

When a native preview is followed by a tool event, appendToolItem() calls settleAssistantForTool(), changing that preview row to streaming: false. This filter therefore preserves the provisional text when a native terminal later supplies an authoritative replacement or explicitly empty final, leaving stale preview content visible beside the final answer. Track native preview ownership separately or remove all preview-owned assistant rows after the latest user boundary, rather than using the mutable streaming flag.

AGENTS.md reference: AGENTS.md:L179-L179

Useful? React with 👍 / 👎.