Invalid JSON payload received - Flash 2.5

[Dayclone]

Branch

dev

Branch (if other)

No response

Version / Tag / Commit

Build (dev): 20260301-1-b62f6e4

Provider(s) Affected

No response

Deployment Method

Binary (exe/executable)

Operating System

Windows 11

Python Version (if running from source)

No response

Bug Description

I get this error trying flash 2.5

[2026-03-16T15:17:10.184021] API error 400: { "error": { "code": 400, "message": "Invalid JSON payload received. Unknown name \"text\" at 'request.system_instruction.parts[0]': Proto field is not repeating, cannot start list.", "status": "INVALID_ARGUMENT", "details": [ { "@type": "type.googleapis.com/google.rpc.BadRequest", "fieldViolations": [ { "field": "request.system_instruction.parts[0]", "description": "Invalid JSON payload received. Unknown name \"text\" at 'request.system_instruction.parts[0]': Proto field is not repeating, cannot start list." } ] } ] } }

Steps to Reproduce

Run it on Flash 2.5 and you should just get this error?

Expected Behavior

It should work.

Actual Behavior

Gives an error

Error and Paylod:
error.log
request_payload.json

Request:
request.json

Error Logs / Messages

Pre-submission Checklist

• I have searched existing issues to ensure this is not a duplicate

[coderabbitai]

⚠️ Possible Duplicate Issue(s)

• can only concatenate str (not \"list\") to str #148

---

🧪 Issue enrichment is currently in open beta.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!

[mirrobot-agent]

@Dayclone Thank you for submitting this issue. I am now beginning my analysis and will report back shortly.

[mirrobot-agent]

Initial Analysis Report

Summary: The error "Invalid JSON payload received. Unknown name "text" at 'request.system_instruction.parts[0]'" is caused by improper handling of system message content in gemini_cli_provider.py.

Issue Validation: Confirmed - This is a valid bug in the Gemini CLI provider implementation.

Reproducibility Assessment: Reproducible - The issue occurs when a client sends a system message with content as a list (multipart format) instead of a plain string, which is valid in the OpenAI API format.

Root Cause Analysis:

In src/rotator_library/providers/gemini_cli_provider.py:653-658, the system message content is extracted and directly wrapped without proper parsing:

system_prompt_content = messages.pop(0).get("content", "")
if system_prompt_content:
    system_instruction = {
        "role": "user",
        "parts": [{"text": system_prompt_content}],
    }

When the client sends a system message like:

{"role": "system", "content": [{"type": "text", "text": "You are helpful"}]}

The code creates:

{"role": "user", "parts": [{"text": [{"type": "text", "text": "You are helpful"}]}]}

This causes the Gemini API to receive a list for the text field instead of a string, triggering the Protocol Buffers error: "Proto field is not repeating, cannot start list."

Comparison with Antigravity Provider:

The antigravity_provider.py:2472-2481 correctly handles this by using _parse_content_parts():

system_content = messages.pop(0).get("content", "")
if system_content:
    new_parts = self._parse_content_parts(system_content, _strip_cache_control=True)
    system_parts.extend(new_parts)

Suggested Labels:

• bug - Clear code defect causing API errors
• gemini_cli - Affects Gemini CLI provider specifically
• needs-fix - Requires code change to resolve

Proposed Next Steps:

1. Immediate Fix: Update _transform_messages() in gemini_cli_provider.py to parse system message content similar to antigravity_provider.py:

   • Add a helper method to parse content (string or list) into Gemini parts format
   • Apply this parsing to system message content before creating system_instruction

2. Code Change Location: src/rotator_library/providers/gemini_cli_provider.py:651-658

3. Testing: Test with both string and list format system messages to ensure compatibility

Missing Information: None - The issue description and error message provide sufficient information to identify the root cause.

This analysis was generated by an AI assistant.

[Mirrowel]

@mirrobot-agent Fix both this issue #149 and issue #148 in a new PR against the dev branch(make sure to pull the dev branch to work on).
Make sure it is robust.

[mirrobot-agent]

@Mirrowel, I'm implementing a robust fix for both issue #149 (Invalid JSON payload) and issue #148 (string/list concatenation error). Both issues stem from improper handling of multipart message content in the Gemini CLI provider. I'll add a proper content parsing method and apply it to system messages and all message types.

[mirrobot-agent]

@Mirrowel, I have successfully implemented and committed the requested changes to fix both issue #149 and issue #148.

Summary

Fixed the "Invalid JSON payload" and "can only concatenate str (not list) to str" errors in the Gemini CLI provider by implementing robust multipart content handling.

Root Cause

Both issues were caused by improper handling of message content when clients send multipart format (list) instead of plain strings. The code was directly wrapping content without parsing, causing the Gemini API to receive lists where it expected strings.

The Fix

1. Added _parse_content_parts() helper method (lines 627-677 in gemini_cli_provider.py):

   • Properly parses both string and list content formats
   • Handles text and image parts
   • Provides consistent error handling

2. Updated message transformation:

   • System messages now use the helper method
   • User messages refactored to use the helper method
   • Assistant messages updated for consistency

3. Ensures compatibility with OpenAI API multipart message format where content can be:

   • Plain string: "Hello"
   • Multipart list: [{"type": "text", "text": "Hello"}, {"type": "image_url", ...}]

Pull Request Created

The changes are now ready for review in:
#150

The PR includes:

• Detailed root cause analysis
• Implementation matching the approach in antigravity_provider.py for consistency
• Comprehensive testing recommendations

This update was generated by an AI assistant.