fix: fixed the problem of failure to call extensionTransport connection tool in ReAct mode

[zzcr]

fix: 修复ReAct模式下调用extensionTransport连接的工具失败问题

Summary by CodeRabbit

• Bug Fixes

  • Fixed transport connection handling to prevent server connection interruptions by refining which connection types are terminated during cleanup operations.

• Chores

  • Version updates: next-remoter (0.0.10-beta.1 → 0.0.10-beta.2) and next-sdk (0.1.15-beta.1 → 0.1.15-beta.2).

_{✏️ Tip: You can customize this high-level summary in your review settings.}

[coderabbitai]

Walkthrough

Version bumps in two package files, with a logic update to AgentModelProvider._closeOneClient method to also exclude ExtensionClientTransport from closure operations alongside existing paired transports, with corresponding comment updates.

Changes

Cohort / File(s)	Summary
Version Bumps packages/next-remoter/package.json, packages/next-sdk/package.json	Version incremented in beta releases: next-remoter (0.0.10-beta.1 → 0.0.10-beta.2) and next-sdk (0.1.15-beta.1 → 0.1.15-beta.2).
Transport Type Exclusion packages/next-sdk/agent/AgentModelProvider.ts	_closeOneClient method updated to exclude ExtensionClientTransport from closure checks alongside InMemoryTransport and MessageChannelTransport. Comments clarified that these three are paired transports that must not be closed to preserve server connections.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~12 minutes

Poem

🐰 A transport so fine, with clients so dear,
No closure shall break what they hold near,
Extension joins kin in the paired transport crew,
Connections preserved—fresh beta version too!

Pre-merge checks and finishing touches

✅ Passed checks (3 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title references extensionTransport connection tool issues in ReAct mode, which aligns with the actual fix to AgentModelProvider.ts that excludes ExtensionClientTransport from being closed.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

✨ Finishing touches

• 📝 Generate docstrings

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch zzc/fix-react-calltool-issue-1213

---

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 and usage tips.}

[coderabbitai]

Actionable comments posted: 0

🧹 Nitpick comments (1)

packages/next-sdk/agent/AgentModelProvider.ts (1)

134-146: Good fix: Correctly prevents closing paired ExtensionClientTransport.

The addition of ExtensionClientTransport to the exclusion list is correct. Like InMemoryTransport and MessageChannelTransport, it's a paired transport where closing the client side would disrupt the server connection (via Chrome extension messaging). This properly fixes the ReAct mode tool calling failure.

Optional: Consider consolidating the two transport exclusion checks

The two separate if blocks (lines 136-141 and 144-146) could be combined for better maintainability:

-      // 如果是 InMemoryTransport 或 MessageChannelTransport，不关闭传输层
-      // 因为它们是配对的，关闭一端会影响另一端（服务端）
-      if (
-        (transport && transport instanceof InMemoryTransport) ||
-        (transport && transport instanceof MessageChannelTransport)
-      ) {
-        return
-      }
-
-      // 因为它们是基于 Chrome 扩展的消息传递机制，关闭会影响服务端的连接
-      if (transport && transport instanceof ExtensionClientTransport) {
-        return
-      }
+      // 如果是配对传输（InMemoryTransport、MessageChannelTransport、ExtensionClientTransport），
+      // 不关闭传输层，因为关闭一端会影响另一端（服务端）的连接
+      if (
+        (transport && transport instanceof InMemoryTransport) ||
+        (transport && transport instanceof MessageChannelTransport) ||
+        (transport && transport instanceof ExtensionClientTransport)
+      ) {
+        return
+      }

📜 Review details

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between a75c17e and fb96e7c.

📒 Files selected for processing (3)

• packages/next-remoter/package.json
• packages/next-sdk/agent/AgentModelProvider.ts
• packages/next-sdk/package.json

🧰 Additional context used

🧬 Code graph analysis (1)

packages/next-sdk/agent/AgentModelProvider.ts (2)

packages/next-sdk/index.ts (1)

• InMemoryTransport (10-10)

packages/next-sdk/transport/ExtensionClientTransport.ts (1)

• ExtensionClientTransport (11-90)

🔇 Additional comments (3)

packages/next-sdk/package.json (1)

3-3: LGTM: Version bump coordinated with the fix.

The beta.2 version increment is appropriate for this bug fix release and aligns with the corresponding version bump in packages/next-remoter/package.json.

packages/next-remoter/package.json (1)

3-3: LGTM: Coordinated version bump.

The beta.2 version increment maintains consistency with the next-sdk package update.

packages/next-sdk/agent/AgentModelProvider.ts (1)

143-146: Listener cleanup bypassed for ExtensionClientTransport prevents proper removal.

The close() method in ExtensionClientTransport correctly calls this._messageListener() to unregister the chrome.runtime.onMessage listener (line 84), but this cleanup is never executed because AgentModelProvider intentionally skips calling close() on these transports (lines 143-146). This causes listeners to accumulate on repeated transport instantiation cycles, creating a memory leak risk.

The current approach assumes transport instances are long-lived, but this assumption should be documented and verified:

1. Confirm if ExtensionClientTransport instances are created/destroyed multiple times during a session
2. If session/lifecycle events trigger new instances, implement explicit cleanup on those boundaries
3. Consider whether the server-connection concerns can be addressed with graceful disconnection instead of skipping cleanup entirely