fix(cursor): pin session id and continue external tool results as userMessageAction

[laerad777]

Summary

• External Cursor wire models now send tool-result continuations as userMessageAction (results stay in history blobs). Native models keep resumeAction.
• Connect x-session-id is pinned to the resolved conversation id so transport rebuilds for the same client thread keep one session identity. Native-exec / background shells use a separate per-transport owner.
• Isolated helper / compaction turns still mint a fresh conversation id, so they do not inherit the parent Connect session.

Rebase source commits: 0046816b3, 4d5850e4d, e3c996586 on current dev (1645bb924).

Verification

• bun run typecheck — clean
• bun run privacy:scan — Privacy scan passed
• bun run test — 12812 pass, 10 skip, 0 fail (826 files)
• Focused: bun test tests/cursor-live-transport.test.ts tests/cursor-adapter.test.ts tests/cursor-blob.test.ts tests/cursor-hardening.test.ts tests/cursor-transport-retry.test.ts — 147 pass

Live Connect probes informed the encoding choice; the unit tests only lock the action case, session-id forwarding, helper isolation, and shell-owner split. They do not claim to reproduce a 45k-token Connect reject.

Checklist

• Scope stays focused and avoids unrelated cleanup.
• Docs or release notes were updated when needed. (no public API / GUI / docs change)
• Security-sensitive changes were reviewed for secrets, auth, and unsafe defaults. (no auth, credentials, workflows, or install-path changes)

Review readiness checklist

• All CI tests are green on my local testing.

• I pushed my PR to the latest dev commit.

• I resolved all correct Codex and CodeRabbit findings.

• My PR is ready for review.

[github-actions]

✅ Deterministic PR hygiene checks passed.

[coderabbitai]

Important

Review skipped

Draft detected.

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: ASSERTIVE

Plan: Pro Plus

Run ID: 877e2500-b66a-45d6-b62a-7fb38d8aa287

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.}

[github-actions]

⏳ DRAFT

• review readiness checklist open (0/4 boxes ticked).

What to do

• Tick all four boxes in the PR description once you're done (currently 0/4).

Review readiness checklist

• ⬜ All CI tests are green on my local testing.
• ⬜ I pushed my PR to the latest dev commit.
• ⬜ I resolved all correct Codex and CodeRabbit findings.
• ⬜ My PR is ready for review.

0/4 boxes ticked.

This PR stays in draft until every box above is ticked.

[lidge-jun]

Keeping this draft open — direction for whoever picks it up (maintainer decision):

What is already on dev (do not re-implement): external tool results continuing as userMessageAction landed with the cursor-call campaign. Rebase first; the half of this diff that duplicates it should disappear in the rebase.

What this PR is still uniquely for: session-id pinning. The adapter still generates a fresh randomUUID() per request, so Cursor sees every turn as a new thread and server-side thread reuse/caching never engages. The remaining work is to pin one stable session id per logical conversation and reuse it across continuations.

Constraints for the implementation:

• Derive the pinned id from the conversation identity the bridge already tracks — do not persist a new store for it.
• The id must survive tool-continuation round-trips but must NOT leak across different logical conversations (two parallel Codex threads must not share one Cursor session).
• Respect the ordering constraint recorded on [Compatibility] Cursor Computer Use / node_repl tool results come back empty or truncated #1866/[Bug] Cursor adapter injects hide-from-user policy and fake user-turn tool-result replay; downstream agents treat it as prompt injection #1992: fix(cursor): stop hide-from-user policy and user-role tool-result replay (#1992) #1997 (assistant-role tool-result replay) merges first; rebase this on top so no user-role replay path returns.
• Add a regression test asserting id stability across a tool continuation and id isolation across two conversations.

After rebase, re-scope the PR description to session pinning only, then it can leave draft.