Two tools with basic schemas and minimal descriptions. create_patch has a 67-char description (below baseline 194 chars); apply_patch_fuzzy has 68 chars. Both lack parameter descriptions entirely, critical gap per pattern:tool-description. Schemas present but parameters undescribed (original_text, modified_text, patch_text, max_drift have no descriptions in the schema). No output schema documented. No error handling guidance. Naming is verb-forward (create_, apply_) but descriptions do not explain WHEN to use each tool or what they return. No mention of idempotency, side effects, or recovery paths.

• Add 100 - 150 character descriptions to each tool. For create_patch: 'Generates a unified diff patch (RFC 3881 format) comparing original and modified code. Use this to create patches for version control or code review. Returns the patch as a string.' For apply_patch_fuzzy: 'Applies a unified diff patch to code with line drift tolerance (fuzzy matching). Handles cases where line numbers have shifted. Returns applied hunks count, conflicts, and patched text.'
• Add descriptions to all parameters in the schema. E.g., original_text: 'The original code text to compare (string, any length)'; modified_text: 'The modified code text (string, any length)'; filename: 'The filename for the patch header (default: file.py, used in diff headers only)'; patch_text: 'A unified diff patch in RFC 3881 format'; max_drift: 'Maximum line number offset to tolerate when matching hunks (integer, 0 - 20, default 5)'.
• Document output schemas. For create_patch, add: 'Returns a string containing the unified diff patch.' For apply_patch_fuzzy, add: 'Returns {status: "success"|"conflicted"|"no_op", applied_hunks: int, total_hunks: int, conflicts: [{hunk_header: string, reason: string}], patched_text: string}'.
• Add error handling guidance to apply_patch_fuzzy description: 'If conflicts occur, the conflicts array lists hunks that could not be matched. Review the hunk headers and consider increasing max_drift or manually inspecting the original code.'
• Clarify that apply_patch_fuzzy modifies state (returns modified code). Add: 'This tool does not persist changes, it returns the patched text. The caller must save or apply the result.' This prevents agents from assuming the patch is automatically committed.