fix: Extract text content from Message objects for agent input

[rodrigosnader]

Summary

Fix agent input format issue where Message objects were passed as LangChain BaseMessage format instead of clean text.

Problem

When Message objects were passed to agents, they were converted to LangChain BaseMessage format which caused input to display as:

(content='text' additional_kwargs={} response_metadata={})

instead of clean text like "text".

Solution

Properly extract the content attribute from LangChain messages to ensure agents receive clean text input.

Changes

• Modified src/lfx/src/lfx/base/agents/agent.py to extract text content from Message objects before passing to agent
• Added proper handling for both Message objects and regular text input

🤖 Generated with Claude Code

Summary by CodeRabbit

• Bug Fixes

  • Improved robustness when providing message-based inputs to agents, preventing errors and inconsistent behavior.
  • Ensures agent inputs are consistently interpreted as text, even when messages lack explicit content.

• Refactor

  • Standardized input normalization to reliably derive text from various input types, enhancing stability and predictability across agent interactions.

[coderabbitai]

Important

Review skipped

Auto incremental reviews are disabled on this repository.

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.

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Walkthrough

Normalizes agent input handling: when input_value is a Message, it’s converted to an LC message and content is extracted (falling back to str). Otherwise, the raw input_value is used. The constructed input_dict now consistently uses input_text to ensure the "input" field is a string or derived content.

Changes

Cohort / File(s)	Summary
Agent input normalization src/lfx/src/lfx/base/agents/agent.py	Added input normalization: convert Message to LC message, extract content or fallback to string; replaced inline input construction with input_text to ensure "input" is consistently string-derived rather than a Message object.

Sequence Diagram(s)

sequenceDiagram
  autonumber
  actor Caller
  participant Agent
  participant LCMessage as LC Message Converter

  Caller->>Agent: run(input_value)
  alt input_value is Message
    Agent->>LCMessage: to_lc_message(input_value)
    LCMessage-->>Agent: lc_message
    Agent->>Agent: extract content or str(lc_message)
  else input_value is not Message
    Agent->>Agent: use input_value directly
  end
  Agent->>Agent: build input_dict with input_text in "input"
  Agent-->>Caller: result

Loading

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Pre-merge checks and finishing touches

❌ Failed checks (1 error, 2 warnings, 1 inconclusive)

Check name	Status	Explanation	Resolution
Test Coverage For New Implementations	❌ Error	I inspected the PR diff and only src/lfx/src/lfx/base/agents/agent.py was modified, with no new or updated test files discovered via the repository search, so the bug fix lacks a regression test that would verify the new Message content extraction behavior.	Please add an automated test that covers passing a Message object to the agent and asserts the normalized text-only input, ensuring the regression is guarded going forward.
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	You can run @coderabbitai generate docstrings to improve docstring coverage.
Test Quality And Coverage	⚠️ Warning	No new or updated tests accompany the change in agent.py, so there is no coverage ensuring that Message objects are converted to plain text or that non-Message inputs still function correctly; manual inspection found no existing tests exercising this normalization path, leaving the updated behavior unverified.	Please add or update tests that pass both Message objects and plain text into the agent input path, asserting that the output is normalized as expected, so the new behavior is covered.
Test File Naming And Structure	❓ Inconclusive	No test files were added or modified in this pull request, and existing tests remain unchanged. Without changes to review, I cannot verify compliance with the required naming patterns, structure, and scenario coverage from the current diff.	Please add or update the relevant tests so their naming, structure, and coverage can be evaluated against the stated criteria, or provide guidance on which existing tests should be reviewed.

✅ 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 succinctly and accurately describes the main change by indicating that text content is extracted from Message objects for agent input, clearly reflecting the purpose of the modifications without extraneous details.
Excessive Mock Usage Warning	✅ Passed	No test files are modified in this pull request, so there is no use of mocks to evaluate and consequently no evidence of excessive mocking related to the proposed changes.

---

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.}

[edwinjosechittilappilly]

I agree with the Fix but curious why it was causing issues for gpt4omini

[edwinjosechittilappilly]

I think we might need to improve to_lc_message also we might need to explore why this error occured I dont think there was any major chnages to to_lc_message. @ogabrielluiz what do you suggest? in this way the input_text is always a string.

[ogabrielluiz]

I think we might need to improve to_lc_message also we might need to explore why this error occured I dont think there was any major chnages to to_lc_message. @ogabrielluiz what do you suggest? in this way the input_text is always a string.

Hey @edwinjosechittilappilly this error occurs because the langchain agent receives a string as input so if you pass something that is not a string it will convert it to string.

[edwinjosechittilappilly]

ok In that case this PR approach seems to be correct.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud