[ENHANCEMENT] apply_diff errors: reduce redundant context; keep best-match details only

[KJ7LNW]

Type

Enhancement

Problem / Value

apply_diff failures return overly verbose errors that repeat the model’s own search content and include large original file snippets. This wastes context and adds noise, making recovery slower.

Context

• Version: v3.23.12
• Provider: Anthropic
• Model: sonnet-3.5
• Current error includes:
  • The search content the model just sent (redundant)
  • Large chunks of original file content (noisy and costly)

Proposal

Streamline the error to include only what helps the next attempt:

• Similarity score and threshold
• “Best Match Found” snippet showing what was actually located
• A brief tip to use read_file for full context when needed

Acceptance Criteria

• Error body excludes repeated search content and large original file dumps
• Includes similarity metrics and the best-match section
• Provides a concise guidance note (e.g., consider read_file for full file)
• Demonstrably reduces tokens used per failure without harming debuggability

Notes

Keep wording consistent with existing tool error styles; prefer brevity and actionable hints over full content dumps.

[roomote]

I am working on this issue to reduce redundant context in apply_diff error messages. I will analyze the current implementation and optimize the error messages to only include the most relevant debugging information.