[BUG] Qwen3-Coder-480B-A35B-Instruct tool-calling failure: Missing value for required parameter 'args'

[steveworks272]

Problem (one or two sentences)

When using the read_file tool, the model fails to provide the required wrapper. It seems to be using the XML format. This results in the error: "roo tried to use read_file without value for required parameter args".

Context (who is affected and when)

Versions of Roo above 3.36.16

Reproduction steps

Use Qwen3-Coder-480B-A35B-Instruct as the provider.
Ask Roo to read a file (e.g., "read package.json").
Observe the model generating <read_file>package.json</read_file> which is rejected by the current schema.

Expected result

use read_file tool to read the file

Actual result

ERROR: roo tried to use read_file without value for required parameter args

Variations tried (optional)

No response

App Version

3.38.3

API Provider (optional)

None

Model Used (optional)

Qwen3-Coder-480B-A35B-Instruct

Roo Code Task Links (optional)

No response

Relevant logs or errors (optional)

[roomote]

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

Root Cause Analysis

The bug occurs because of a mismatch between the tool protocol mode and tool availability:

1. Protocol Resolution: resolveToolProtocol() now always returns "native" for new tasks, deprecating XML
2. Tool Inclusion Logic: In Task.ts:3909, tools are only sent when:

   const shouldIncludeTools = taskProtocol === TOOL_PROTOCOL.NATIVE && (modelInfo.supportsNativeTools ?? false)

3. IO Intelligence models don't have supportsNativeTools: true set in their model info
4. So when using Qwen3-Coder-480B via IO Intelligence:
   • Task is in "native" mode (always now)
   • supportsNativeTools is undefined, defaults to false
   • shouldIncludeTools = false, so native tool schemas are NOT sent
   • XML parser is disabled (because we're in native mode)
   • Model sees XML tool instructions in system prompt, outputs XML in text
   • XML in text is NOT parsed, tools fail to execute properly

Proposed Fix

Option A (Recommended): Add supportsNativeTools: true to IO Intelligence model info

Since IO Intelligence uses OpenAI-compatible API which supports function calling, we should mark these models as supporting native tools. This fix involves:

1. Update src/api/providers/fetchers/io-intelligence.ts to set supportsNativeTools: true when fetching models
2. Update src/api/providers/io-intelligence.ts to include supportsNativeTools: true in the default model info

This is the minimal change that aligns IO Intelligence with other OpenAI-compatible providers.

Option B (Alternative): Automatic fallback to XML when native tools aren't supported

If the model doesn't support native tools, automatically use XML parsing. This would be a more defensive fix but changes the protocol resolution logic.

---

I recommend Option A as it's the targeted fix for IO Intelligence.

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

[LeonYoah]

How to temporarily fix

[hannesrudolph]

Thanks. This is a bug report, but it is missing the provider details needed to reproduce. Please more carefully fill out the bug report including your Roo version and provider (not just model).