Can not write files

[dctr]

Description

I am consistently seeing write failures when trying to write new files. (Small patches usually work.)

Most of the time, I see the tool just having {} arguments, and the return is `Write Failed: Tool execution aborted.

One out of the 10+ attempts looked like this:

The write tool was called with invalid arguments: [
  {
    "expected": "string",
    "code": "invalid_type",
    "path": [
      "content"
    ],
    "message": "Invalid input: expected string, received undefined"
  },
  {
    "expected": "string",
    "code": "invalid_type",
    "path": [
      "filePath"
    ],
    "message": "Invalid input: expected string, received undefined"
  }
].
Please rewrite the input so it satisfies the expected schema.

Plugins

None

OpenCode version

Desktop, Version 1.2.27

Steps to reproduce

1. Tell AI to come up with "something", and have it write it to a file.
2. AI uses write tool call
3. ???
4. Repeats several times and fails

Screenshot and/or share link

No response

Operating System

macOS 26.3.1

Terminal

Desktop app, but also happens on iTerm

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• Write tool called with invalid parameters #18131: Identical error message — write tool called with invalid parameters (Invalid input: expected string, received undefined for both content and filePath), same version 1.2.27
• [BUG] Write tool continously getting aborted #11079: Write tool continuously getting aborted (similar symptom of write failures)

If your issue is the same as #18131, please consider adding your details there instead.

[ai-nurmamat]

Root Cause Analysis

This issue appears to be related to how the write tool arguments are being processed/passed through the AI model response pipeline.

Error Analysis

The error indicates:

For both and fields - this suggests the arguments are being lost or undefined somewhere in the tool call chain.

Likely Root Causes

1. Tool Call Argument Parsing Issue
The tool is likely implemented as part of the edit/apply system. When the AI model generates a write operation, the arguments may not be properly extracted or validated before being passed to the tool execution layer.

2. Streaming Response Handling
If using streaming responses from the LLM, there could be a race condition where tool calls are parsed before complete arguments are received.

3. Schema Validation Timing
The error shows arguments are - this typically happens when:

• The tool call arguments object is created before all fields are populated
• There's a schema mismatch between what the model outputs and what the tool expects

Recommended Investigation

1. Check how tool is defined in the tool registry
2. Look at the tool argument extraction/validation logic
3. Verify the schema definition matches what the LLM is being told to output

Possible Fix

Ensure the tool schema explicitly marks fields as required and add validation before executing the tool to provide better error messages.

[dctr]

To clarify, I am not running a custom model. This happens with Claude Sonnet/Opus 4.6

[Taudris]

Still happening on 1.3.13 with Opus 4.6 on Copilot.

It also completely stops the agent, which is damaging to productivity.