🐛 Avoid displaying Gemini tool calls

[abrugaro]

Resolves #1129

Summary by CodeRabbit

• Bug Fixes
  • Improved extraction and processing of text content from structured AI-generated streaming responses, ensuring more reliable handling of streamed messages.
  • Reduces serialized artifacts and missing fragments in streams for smoother, more accurate text display during live AI replies.
  • Users will see fewer formatting glitches and better continuity in streamed reply text.

[snyk-io]

⛔ Snyk checks have failed. 1 issues have been found so far.

Status	Scanner	Critical	High	Medium	Low	Total (1)
⛔	Open Source Security	0	0	1	0	1 issues

💻 Catch issues earlier using the plugins for VS Code, JetBrains IDEs, Visual Studio, and Eclipse.

[coderabbitai]

📝 Walkthrough

Walkthrough

The handler for KaiWorkflowMessageType.LLMResponseChunk in processMessage now treats chunk.content arrays by filtering elements with type === "text" and string text values, then concatenating those text fields into a single string. Previously non-string array items were serialized with JSON.stringify; now only textual parts are extracted and joined, aligning array-based streaming content with existing string handling.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~2 minutes

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 inconclusive)

Check name	Status	Explanation	Resolution
Description check	❓ Inconclusive	The PR description only contains 'Resolves #1129' without providing details about the change, implementation approach, or reasoning beyond referencing the linked issue.	Expand the description to explain what was changed and how it addresses the issue (e.g., the array filtering logic added to handle text-only content).

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The PR title ':bug: Avoid displaying Gemini tool calls' clearly describes the main change and follows the repository's emoji prefix convention for patch fixes.
Linked Issues check	✅ Passed	The code change filters array content to extract only text parts, which directly addresses the issue of displaying Gemini tool calls by suppressing non-text debug information.
Out of Scope Changes check	✅ Passed	The modification to processMessageByType is narrowly scoped to filtering LLMResponseChunk array content for text parts only, directly supporting the issue's objective of suppressing tool call display.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing touches

• 📝 Generate docstrings

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment

---

No actionable comments were generated in the recent review. 🎉

Tip

Issue Planner is now in beta. Read the docs and try it out! Share your feedback on Discord.

---

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: 1

🤖 Fix all issues with AI agents

In `@vscode/core/src/utilities/ModifiedFiles/processMessage.ts`:
- Around line 304-308: In processMessage, when handling the
Array.isArray(chunk.content) branch, guard against arrays that contain only
non-text tool-call parts: compute the filtered list (e.g., filteredTexts =
chunk.content.filter(...).map(...)) and if filteredTexts.length === 0 then skip
creating/using content (e.g., return early or continue so no empty string is
propagated), otherwise set content = filteredTexts.join(""); ensure this uses
the existing chunk.content and content variables so downstream code (message
append logic) never receives an empty "" chat message.