fix(agent): skip reasoning-only assistant messages to prevent empty m…

[Cprakhar]

What

Skip adding reasoning-only assistant messages to chat state in Agent.parseUiStream.

Why

Fixes #705 - empty messages were being created when assistant only had reasoning/step-start parts, leading to "Agent failed: Bad Request [ce]" errors.

How

Added a guard that filters out assistant messages containing only reasoning or step-start parts before adding to karton state.

Testing

• Manual testing: verify no empty messages appear in chat
• Changeset included for version bump

Summary by CodeRabbit

• Bug Fixes
  • Fixed a Bad Request error that occurred when internal reasoning-only messages were added to the chat stream, preventing empty messages from causing failures.
  • Improved message clarity by automatically filtering out internal reasoning and processing step information from displayed messages, keeping the chat history focused on user-relevant content.

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

[cla-bot]

We require contributors to sign our Contributor License Agreement, and we don"t have @Cprakhar on file. In order for us to review and merge your code, please follow our CLA guidelines.

[vercel]

@Cprakhar is attempting to deploy a commit to the stagewise Team on Vercel.

A member of the Team first needs to authorize it.

[coderabbitai]

Walkthrough

A new changeset entry documents a patch for @stagewise/agent-client, and filtering logic is added to the UI stream parser to skip assistant messages containing only non-user-visible reasoning or step-start parts, preventing empty messages from causing Bad Request errors.

Changes

Cohort / File(s)	Summary
Changeset documentation \.changeset/itchy-files-doubt.md	New changeset documenting patch release for @stagewise/agent-client with note about skipping reasoning-only messages to avoid empty message errors.
Message filtering logic agent/client/src/Agent.ts	Adds filtering in UI stream parser to skip assistant messages where all parts are either 'reasoning' or 'step-start' type, preventing internal reasoning fragments from surfacing in user-visible chat history.

Sequence Diagram

sequenceDiagram
    participant Stream as UI Stream
    participant Parser as parseUiStream
    participant Filter as Message Filter
    participant History as Chat History

    Stream->>Parser: Emit assistant message with parts
    Parser->>Filter: Check message parts
    
    alt All parts are reasoning or step-start
        Filter-->>Parser: Skip message
        Note over Parser,History: Message filtered out
    else Contains user-visible content
        Filter->>History: Add message to chat
        Note over History: Message surfaced to user
    end

Loading

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

• Review the filtering logic in Agent.ts to confirm it correctly identifies and skips reasoning-only messages
• Verify the changeset accurately reflects the fix for issue #705

Possibly related PRs

• feat(agent): enhance reasoning capabilities and improve UI message handling #674: Implements identical message filtering for reasoning/step-start-only assistant messages in the UI stream parser.
• fix(agent): skip assistant messages that contain only reasoning parts #678: Implements the same code change to skip assistant messages composed only of reasoning or step-start parts.
• Fix fetch failed error #572: Modifies UI streaming and message parsing in Agent.ts; this PR complements that work by adding upstream filtering.

Suggested labels

cla-signed

Poem

🐰 Through reasoning streams and steps untold,
We filter noise with logic bold,
No empty whispers shall pass through,
Only wisdom users view! ✨

Pre-merge checks and finishing touches

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly summarizes the main change: skipping reasoning-only assistant messages to prevent empty messages, which directly addresses the linked issue #705.
Linked Issues check	✅ Passed	The PR successfully implements filtering of assistant messages containing only reasoning or step-start parts, directly addressing the core requirement from issue #705 to prevent empty messages that cause Bad Request errors.
Out of Scope Changes check	✅ Passed	All changes are scoped to the stated objective: the Agent.ts modification filters reasoning-only messages, and the changeset documents the patch release—both directly addressing issue #705.
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

Tip

📝 Customizable high-level summaries are now available in beta!

You can now customize how CodeRabbit generates the high-level summary in your pull requests — including its content, structure, tone, and formatting.

• Provide your own instructions using the high_level_summary_instructions setting.
• Format the summary however you like (bullet lists, tables, multi-section layouts, contributor stats, etc.).
• Use high_level_summary_in_walkthrough to move the summary from the description to the walkthrough section.

Example instruction:

"Divide the high-level summary into five sections:

1. 📝 Description — Summarize the main change in 50–60 words, explaining what was done.
2. 📓 References — List relevant issues, discussions, documentation, or related PRs.
3. 📦 Dependencies & Requirements — Mention any new/updated dependencies, environment variable changes, or configuration updates.
4. 📊 Contributor Summary — Include a Markdown table showing contributions:
   | Contributor | Lines Added | Lines Removed | Files Changed |
5. ✔️ Additional Notes — Add any extra reviewer context.
   Keep each section concise (under 200 words) and use bullet or numbered lists for clarity."

Note: This feature is currently in beta for Pro-tier users, and pricing will be announced later.

---

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)

agent/client/src/Agent.ts (1)

680-690: LGTM! Logic correctly implements the fix for the reported issue.

The filtering logic properly prevents assistant messages containing only reasoning or step-start parts from being added to the chat state. The condition ordering is safe—parts?.length > 0 ensures parts is defined before calling .every().

Optional enhancement: Consider also filtering assistant messages with empty or undefined parts arrays to be more defensive:

 // Skip assistant messages that only contain non-user-visible reasoning or step-start parts
 if (
   uiMessage.role === 'assistant' &&
-  uiMessage.parts?.length > 0 &&
-  uiMessage.parts.every(
-    (part) =>
-      part.type === 'reasoning' || part.type === 'step-start',
-  )
+  (!uiMessage.parts ||
+    uiMessage.parts.length === 0 ||
+    uiMessage.parts.every(
+      (part) =>
+        part.type === 'reasoning' || part.type === 'step-start',
+    ))
 ) {
   continue;
 }

This would also catch any assistant messages with no content, though the current implementation correctly addresses the specific bug reported in #705.

📜 Review details

Configuration used: CodeRabbit UI

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 589104e and c41c9c4.

📒 Files selected for processing (2)

• .changeset/itchy-files-doubt.md (1 hunks)
• agent/client/src/Agent.ts (1 hunks)

🧰 Additional context used

📓 Path-based instructions (1)

**/*.{ts,tsx}

📄 CodeRabbit inference engine (CLAUDE.md)

NEVER use any. NEVER use unknown unless absolutely necessary.

Files:

• agent/client/src/Agent.ts

🧠 Learnings (1)

📚 Learning: 2025-08-17T19:03:32.418Z

Learnt from: CR
Repo: stagewise-io/stagewise PR: 0
File: toolbar/core/CLAUDE.md:0-0
Timestamp: 2025-08-17T19:03:32.418Z
Learning: Applies to toolbar/core/src/hooks/agent/**/*.{ts,tsx} : Place agent connectivity hooks under `src/hooks/agent`, with individual interface capabilities covered by individual hooks

Applied to files:

• agent/client/src/Agent.ts

🔇 Additional comments (1)

.changeset/itchy-files-doubt.md (1)

1-5: LGTM!

The changeset correctly documents the patch and provides a clear description of the fix that aligns with the PR objectives.

[glenntws]

@Cprakhar thanks for your contribution! We would be ready to approve this once everything checks out and the CLA is signed!

@juliangoetze should be an easy merge i guess?