[BUG] attempt_completion loop (Grok fast)

[clover-supply]

Problem (one or two sentences)

Grok is repeatedly getting rejected/ignored by the system on task completions despite clearly callling it. But it completes after few autoretries.

Context (who is affected and when)

Grok fast in VS code extension

Reproduction steps

attempt to finish a task to do list with Grok

Expected result

no loop

Actual result

loop

Variations tried (optional)

No response

App Version

Version: 3.39.3 (85d253c)

API Provider (optional)

None

Model Used (optional)

Grok code fast 1

Roo Code Task Links (optional)

No response

Relevant logs or errors (optional)

[daniel-lxs]

Hey can you follow https://docs.roocode.com/reporting-errors and share more information about this failing task?

[roomote]

I explored the codebase to understand potential causes for this attempt_completion loop with Grok. Here is my analysis:

Potential Causes

The attempt_completion loop could be caused by one of these scenarios:

1. Tool call not being detected: The system checks assistantMessageContent for tool_use blocks. If Grok's tool call is not being properly parsed/recognized, the system would show "MODEL_NO_TOOLS_USED" error and ask the model to retry.

2. Native tool call parsing issue: Grok uses the native tool call protocol via NativeToolCallParser. If the streaming chunks are malformed or missing expected fields (like id), the tool call might not be properly initialized.

3. finish_reason mismatch: The system expects finish_reason: "tool_calls" to finalize tool calls. If Grok sends a different finish_reason, tool calls are only finalized at stream end via finalizeRawChunks().

Information That Would Help

Once the user provides logs as @daniel-lxs requested, specifically look for:

• Whether "MODEL_NO_TOOLS_USED" errors appear in the logs
• The raw API response chunks from Grok (especially the finish_reason value)
• Whether tool_call_start and tool_call_end events are being emitted

Additionally:

• Are you using Grok via the xAI provider directly, or through Roo Code Cloud?
• Does this happen only with attempt_completion, or with other tools as well?

Please tag @roomote when you provide this information, and I'll investigate further.

[clover-supply]

Update, this appears connected to the "power steering" experimental feature.

edit: Grok via roo. Only seen it so far on attempt_completion and once the system called for the tool when not needed.

[clover-supply]

@roomote

system complained about no tool use despite grok using attempt completion
roo-diagnostics-85ddb9ee-1768334652330.json

system called for tool when not needed
roo-diagnostics-85ddb9ee-1768334491436.json

unknown
roo-diagnostics-85ddb9ee-1768333777996.json

Hey can you follow https://docs.roocode.com/reporting-errors and share more information about this failing task?

Date/time: 2026-01-13T19:40:31.083Z
Extension version: 3.39.3
Provider: roo
Model: xai/grok-code-fast-1

The model provided text/reasoning but did not call any of the required tools. This usually indicates the model misunderstood the task or is having difficulty determining which tool to use. The model has been automatically prompted to retry with proper tool usage.

the same one three times with