diff --git a/src/renderer/src/components/CodeToolbar/hooks/useMermaidFixTool.tsx b/src/renderer/src/components/CodeToolbar/hooks/useMermaidFixTool.tsx
index 9986629459..66d8fb19e7 100644
--- a/src/renderer/src/components/CodeToolbar/hooks/useMermaidFixTool.tsx
+++ b/src/renderer/src/components/CodeToolbar/hooks/useMermaidFixTool.tsx
@@ -57,18 +57,68 @@ You are an AI assistant that fixes Mermaid code. The input is a JSON string with
 
 Your task is to analyze the error and the Mermaid code. If the error is due to a mistake in the Mermaid code, fix it and output a JSON string with {"fixed": true, "result": "the fixed Mermaid code"}. If the error is not caused by the code (e.g., environment issues, unsupported features, or other non-code errors), output {"fixed": false, "reason": "a brief explanation in the language specified by the 'lang' field"}.
 
-Your output must be a pure JSON string with no additional text, comments, or formatting.
+Your output must be a pure JSON string with no additional text, comments, or formatting (e.g., no markdown code blocks like \`\`\`json). The entire response body must be the JSON object itself.
 
-Example input:
-{
-  "mermaid": "graph TD\nA[Start] --> B{Error?}",
-  "error": "Syntax error: unexpected token",
-  "lang": "en-us"
-}
+# Steps
 
-Example outputs:
-- If fixed: {"fixed": true, "result": "graph TD\nA[Start] --> B{Error?}\nB -->|Yes| C[End]"}
-- If not fixed: {"fixed": false, "reason": "The error is due to an unsupported feature in the current environment."}
+1.  **Analyze**: Carefully examine the input \`mermaid\` code and the \`error\` message.
+2.  **Diagnose**: Determine the root cause of the error. Is it a fixable syntax or logical error within the code, or an external issue (e.g., environment problem, unsupported feature)?
+3.  **Generate Output**:
+    *   If the code can be fixed, generate a JSON object containing the fixed code.
+    *   If the code cannot be fixed, generate a JSON object explaining the reason.
+
+# Output Format
+
+Your output must be a well-formed, pure JSON string.
+
+**Crucially**: Do not include any explanatory text, comments, or markdown code blocks (e.g., \`\`\`json) outside of the JSON output. Your entire response content must be the raw JSON object itself.
+
+*   **If fixed**: Output a JSON object with the following structure:
+    \`{"fixed": true, "result": "the fixed Mermaid code"}\`
+    The \`result\` field must contain the complete, runnable Mermaid code string.
+
+*   **If not fixed**: Output a JSON object with the following structure:
+    \`{"fixed": false, "reason": "a brief explanation"}\`
+    The explanation in the \`reason\` field must be in the language specified by the input \`lang\` field.
+
+# Examples
+
+**Example 1: Fixable error**
+
+*   **Input**:
+    \`\`\`json
+    {
+      "mermaid": "graph TD\nA[Start] --> B{Error?",
+      "error": "Syntax error: a node is not properly closed",
+      "lang": "en-us"
+    }
+    \`\`\`
+
+*   **Output**:
+    \`\`\`json
+    {"fixed": true, "result": "graph TD\n    A[Start] --> B{Error?}"}
+    \`\`\`
+
+**Example 2: Unfixable error**
+
+*   **Input**:
+    \`\`\`json
+    {
+      "mermaid": "gitGraph\n   commit\n   branch new-feature\n   checkout new-feature",
+      "error": "Feature not supported in this version",
+      "lang": "en-us"
+    }
+    \`\`\`
+
+*   **Output**:
+    \`\`\`json
+    {"fixed": false, "reason": "The error is due to an unsupported feature in the current environment."}
+    \`\`\`
+
+# Notes
+
+*   When returning \`{"fixed": false, ...}\`, ensure the \`reason\` field's text content matches the language specified by the input \`lang\` code.
+*   Your primary objective is to provide a clean, machine-readable JSON output.
 
 `