refactor: simplify apply_diff XML schema

[daniel-lxs]

Summary

This PR simplifies the XML schema for the apply_diff tool by removing unnecessary complexity and improving clarity.

Changes

1. Removed Backticks from XML Content

• Before: XML content blocks contained triple backticks (```) mixing markdown syntax with XML
• After: Clean XML without backticks, making parsing more straightforward

2. Eliminated Redundant args Wrapper

• Before: <args><file>...</file></args>
• After: <file>...</file> (files are direct children)

3. Removed Duplicate start_line Specification

• Before: start_line could be specified both as XML element and within content
• After: Only use in-content markers (:start_line:) when needed

4. Simplified Diff Structure

• Before: <diff><content>...</content></diff>
• After: <diff>...</diff> (content directly in diff element)

Example of New Schema

<apply_diff>
<file>
  <path>example.py</path>
  <diff>
<<<<<<< SEARCH
def old_function():
    return 'old'
=======
def new_function():
    return 'new'
>>>>>>> REPLACE
  </diff>
</file>
</apply_diff>

Benefits

1. Cleaner Schema: No mixing of markdown and XML syntax
2. Simpler Parsing: Fewer nested elements to parse
3. Less Redundancy: Single way to specify metadata
4. Better Error Messages: Clearer structure makes errors easier to understand
5. Backward Compatibility: Legacy format still supported

Testing

• Updated parsing logic to handle new schema
• Maintained backward compatibility with old format
• Updated tool descriptions and examples
• Manual testing with various diff scenarios

Breaking Changes

While backward compatibility is maintained, this changes the recommended XML schema format for the apply_diff tool. Documentation and examples have been updated accordingly.

[roomote]

Thank you for your contribution! I've reviewed the changes and have some suggestions for improvement. Overall, this is a good refactoring that simplifies the XML schema as intended. The backward compatibility implementation looks solid, and the cleaner schema will make the tool easier to use.

[roomote]

Is this intentional? The code extracts :start_line: from the diff content here, but the documentation in multi-file-search-replace.ts no longer mentions this format. Should we update the documentation to clarify that :start_line: can still be used within the diff content for cases where it's needed?

[roomote]

The error message shows the simplified structure, which is great! However, would it be helpful to also mention that the legacy format is still supported for backward compatibility? This could help users who are migrating from the old format.

[roomote]

Consider adding a type guard or validation here to ensure diff is actually a string in the simplified format case. While String(diff) works, explicit type checking would make the code more robust:

Suggested change

	diffContent = String(diff)
	// New simplified format - diff is the content string directly
	if (typeof diff !== 'string') {
	throw new Error('Invalid diff format: expected string for simplified format');
	}
	diffContent = diff;

[roomote]

Since this maintains backward compatibility but changes the recommended format, would it be helpful to add a deprecation notice when the legacy format is detected? This could guide users to migrate to the cleaner schema over time. For example:

Suggested change

	diffContent = diff.content
	// Legacy format with content wrapper
	console.warn('[Deprecation] The <content> wrapper in apply_diff is deprecated. Please use the simplified format with diff content directly.');
	diffContent = diff.content

[daniel-lxs]

This broke the tool

[hannesrudolph]

This is the best solution! 👍

After investigating the issue thoroughly, this PR provides the cleanest fix. By removing the backticks entirely from the XML examples, we eliminate the root cause of the parsing errors without needing:

• CDATA sections (as in PR fix: use CDATA sections in XML examples to prevent parser errors (#4852) #6811)
• Fallback parsers (as in PR fix: prevent apply_diff hanging on XML parsing with external interference (#4852) #6774)
• Any workarounds

The simplified schema is much cleaner and more maintainable. This should be the preferred solution for fixing issue #4852.

Summary of approaches:

• PR fix: prevent apply_diff hanging on XML parsing with external interference (#4852) #6774: Added complex fallback parser (based on incorrect diagnosis of xml2js interference)
• PR fix: use CDATA sections in XML examples to prevent parser errors (#4852) #6811: Used CDATA sections to escape backticks (works but adds XML complexity)
• PR refactor: simplify apply_diff XML schema #6812: Removed backticks entirely (cleanest, simplest solution) ✅

I recommend merging this PR and closing the others.

[Avinm]

Not sure if it's related, but I found that RooCode fails on this prompt:

Can you create test.md with the below lines verbatim?
```mermaid
```

It just writes an empty file. Would this fix it?