[BUG] API Request Failed – 400: No tool output found for function call

[thebrainteasehub-png]

Problem (one or two sentences)

’I am facing a major blocking issue that makes the tool practically unusable.

Issue:
Every time I try to run a task (even moderately complex ones), I frequently get the following error:

“API Request Failed – 400: No tool output found for function call”

This happens repeatedly, even after retrying, and results in zero output. The task simply fails without any recovery.

Impact:

• No code is generated at all
• Wastes significant time waiting for long-running requests
• Forces restarting tasks repeatedly
• Breaks development workflow completely

Overall:
This is a critical issue that needs urgent attention. The current behavior severely affects usability, especially for paid users expecting reliable performance.

Context (who is affected and when)

Observations:

• Occurs more often with larger prompts or file references (e.g., modifying a Python dashboard file)
• Retrying the same task does not fix the issue
• Starting a new task sometimes works, but not reliably
• Feels like a tool execution or response synchronization failure rather than a prompt issue

Reproduction steps

• Occurs more often with larger prompts or file references (e.g., modifying a Python dashboard file)
• Retrying the same task does not fix the issue
• Starting a new task sometimes works, but not reliably
• Feels like a tool execution or response synchronization failure rather than a prompt issue

Expected result

Expected Behavior: * Either the tool should complete execution successfully * Or provide a proper fallback/error recovery instead of failing silently

Actual result

[BUG] API Request Failed – 400: No tool output found for function call

Variations tried (optional)

No response

App Version

3.53.0

API Provider (optional)

None

Model Used (optional)

5.4 Pro(Open AI)

Roo Code Task Links (optional)

No response

Relevant logs or errors (optional)

[roomote-v0]

I've analyzed this issue and here's my proposed implementation plan:

Root Cause

The error "No tool output found for function call" (HTTP 400) comes from OpenAI's Responses API when it receives a function_call item in the conversation input without a matching function_call_output item. I found two bugs in the OpenAI Native provider that can cause this:

Bug 1: Ordering issue in formatFullConversation() (src/api/providers/openai-native.ts)

When a user message contains both tool results and text content (e.g., user feedback alongside tool execution results), the code currently emits user text content before function_call_output items:

function_call (from assistant)     <-- OpenAI sees this tool call
user message (text feedback)       <-- ...then a user message
function_call_output               <-- ...then the tool output (too late!)

OpenAI's Responses API expects function_call_output to appear immediately after its corresponding function_call. Inserting a user message between them breaks the pairing. This occurs in real usage when Roo Code adds user feedback text alongside tool results in the same message, which happens frequently with larger/more complex prompts (matching the reporter's observations).

Bug 2: SDK error fallback retries the same failing request (src/api/providers/openai-native.ts)

In executeRequest(), when the SDK throws any error (including 400 validation errors), the code falls back to makeResponsesApiRequest() which sends the exact same malformed request via manual fetch. This doubles the latency on failure (the user waits for the same request to fail twice) without any recovery.

Proposed Fix

1. Fix ordering in formatFullConversation(): Swap the order so function_call_output items are emitted before user text content. This ensures every function_call is immediately followed by its function_call_output.

2. Fix error fallback in executeRequest(): Only fall back to manual SSE for non-4xx errors (network issues, SDK streaming issues, etc.). For 4xx errors (client validation errors like the one in this issue), re-throw immediately since retrying the same payload won't help.

3. Add tests: Add unit tests for formatFullConversation() to verify correct ordering of function_call_output before user content, and for the error fallback behavior.

Please mention @roomote in a comment saying that you approve this plan, and I'll proceed with the implementation.