feat(agent-tracing): collect assistant messages into root run outputs

[jerryliang64]

Summary

• Automatically collect assistant text responses during message processing into outputs.messages
• Aligns with LangGraph convention where root run outputs contains business data (messages)

Changes

• core/agent-tracing/src/ClaudeAgentTracer.ts
  • Add outputMessages array to Trace class to accumulate assistant text
  • Collect text from assistant messages in handleAssistant
  • Include messages in root run outputs in handleResult
• core/agent-tracing/test/ClaudeAgentTracer.test.ts
  • New test: multi-turn assistant text collected into outputs.messages
  • New test: empty messages array when no assistant text

Root Run outputs (before → after)

 outputs: {
+  messages: [{ role: 'assistant', content: '...' }, ...],
   result: '...',
   is_error: false,
   num_turns: 1,
   llmOutput: { promptTokens, completionTokens, ... },
 }

Test plan

• All 58 tests pass (56 existing + 2 new)
• Verified locally in chair-sandbox-ai-use project

🤖 Generated with Claude Code

Summary by CodeRabbit

• New Features

  • Assistant text content is now collected per trace and exposed in trace outputs as a messages array.
  • Mixed content blocks (text and tool-use) are preserved in the recorded messages.

• Tests

  • Added tests verifying message collection for multi-part assistant outputs and the empty messages case when no assistant text appears.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: b5b68feb-8cd5-4083-bcfc-909ade4a0223

📥 Commits

Reviewing files that changed from the base of the PR and between 78685ea and b5df79a.

📒 Files selected for processing (2)

• core/agent-tracing/src/ClaudeAgentTracer.ts
• core/agent-tracing/test/ClaudeAgentTracer.test.ts

✅ Files skipped from review due to trivial changes (1)

• core/agent-tracing/test/ClaudeAgentTracer.test.ts

🚧 Files skipped from review as they are similar to previous changes (1)

• core/agent-tracing/src/ClaudeAgentTracer.ts

---

📝 Walkthrough

Walkthrough

Collect assistant content blocks per-trace by adding an outputMessages array to Trace. On assistant messages with content blocks, append { role: 'assistant', content }. On final result, attach messages: this.outputMessages to the root run's outputs while preserving existing fields and behavior.

Changes

Cohort / File(s)	Summary
Core Implementation core/agent-tracing/src/ClaudeAgentTracer.ts	Added outputMessages to Trace; append assistant content blocks as { role: 'assistant', content } on assistant messages; include outputs.messages with collected entries on root run result.
Test Suite core/agent-tracing/test/ClaudeAgentTracer.test.ts	Added tests under Trace outputs.messages in root run validating collected assistant message entries (mixed content and text-only) and empty-array behavior when no assistant content exists.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related PRs

• feat(agent-tracing): rename TraceSession to Trace and support inputs in createTrace #420: Modifies tracing/root-run output handling and finalization logic, related to root run outputs.
• feat: add @eggjs/agent-tracing package for AI agent tracing #412: Previously adjusted ClaudeAgentTracer behavior; touches the same tracer implementation and message-handling flows.

Suggested reviewers

• akitaSummer

Poem

"I hop through traces, soft and light,
Gathering assistant words in sight,
Little blocks tucked in a row,
Saved for the root when runs must close,
A happy rabbit with logs aglow." 🐇✨

🚥 Pre-merge checks | ✅ 3

✅ 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 accurately and concisely describes the main change: collecting assistant messages into root run outputs, which aligns with the core functionality added to ClaudeAgentTracer.ts.
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

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch feat/claude-tracer-output-messages

---

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.}

[gemini-code-assist]

Code Review

This pull request introduces the collection of assistant text messages into the outputs.messages field of the root run within the ClaudeAgentTracer. This is achieved by tracking text content during message processing and updating the root run's outputs upon completion. Corresponding tests were added to ensure correct message aggregation. Feedback suggests improving type safety by narrowing the role property to a literal type and using type guards instead of type assertions when filtering content.