新增: 工具工作流补全(diff_files工具 + 技能清单注入 + 工作流进度共享 + human端到端测试 + 知识库MCP工具)

src-tauri/src/commands/ai/tools/file.rs

@@ -259,6 +259,46 @@ pub fn register(
         }
     );
 
+    // ── diff_files (Low 只读, 任意两文件对比) ──
+    // 补 git_diff 缺口:git_diff 只比仓库内版本,本工具对比任意两文件(绝对路径),
+    // 复用 generate_diff(LCS unified diff)生成差异,只读无副作用。
+    declare_tool!(
+        registry,
+        allowed_dirs: Arc<RwLock<AllowedDirs>>,
+        "diff_files",
+        "对比任意两个文件的差异，返回 unified diff（+/- 行前缀，基于 LCS）。参数 path_a/path_b 为两文件绝对路径，均须在授权目录内。只读无副作用，用于两文件/两版本内容对比（git_diff 仅仓库内，本工具补任意两文件缺口）",
+        RiskLevel::Low,
+        schema: object_schema(vec![("path_a", "string", true), ("path_b", "string", true)]),
+        args => {
+            let snap = allowed_dirs.read().await.clone();
+            let a_resolved = resolve_workspace_path_with_allowed(
+                args["path_a"].as_str().ok_or_else(|| anyhow::anyhow!("缺少 path_a 参数"))?,
+                &snap,
+            )?;
+            let b_resolved = resolve_workspace_path_with_allowed(
+                args["path_b"].as_str().ok_or_else(|| anyhow::anyhow!("缺少 path_b 参数"))?,
+                &snap,
+            )?;
+            let a_path = a_resolved.to_str().ok_or_else(|| anyhow::anyhow!("path_a 含非法字符"))?.to_string();
+            let b_path = b_resolved.to_str().ok_or_else(|| anyhow::anyhow!("path_b 含非法字符"))?.to_string();
+            if a_path == b_path {
+                anyhow::bail!("两个路径相同，无需对比: {}", a_path);
+            }
+            // 读取两文件文本(1MB 上限 + 二进制拦截,解码对齐 read_file 的 UTF-16 BOM 优先)
+            let (content_a, size_a) = read_text_file(&a_path).await?;
+            let (content_b, size_b) = read_text_file(&b_path).await?;
+            let diff = generate_diff(&content_a, &content_b);
+            Ok(serde_json::json!({
+                "path_a": a_path,
+                "path_b": b_path,
+                "size_a": size_a,
+                "size_b": size_b,
+                "identical": content_a == content_b,
+                "diff": diff,
+            }))
+        }
+    );
+
     // ── list_directory (Low 只读) ──
     declare_tool!(
         registry,
@@ -1232,6 +1272,38 @@ pub fn register(
     );
 }
 
+// ============================================================
+// diff_files 辅助 — 读取文本文件内容(两文件对比共用)
+// ============================================================
+
+/// 读取文本文件内容，返回 (内容, 字节数)。
+///
+/// diff_files 两文件对比共用：1MB 上限 + 二进制拦截 + UTF-16 BOM 优先解码
+/// （复用 decode_bytes_to_string，对齐 read_file/patch_file 的读取语义，单真相源）。
+async fn read_text_file(path: &str) -> anyhow::Result<(String, u64)> {
+    use tokio::fs::File;
+    use tokio::io::AsyncReadExt;
+    let mut file = File::open(path).await
+        .map_err(|e| {
+            if e.kind() == std::io::ErrorKind::NotFound {
+                anyhow::anyhow!("文件不存在: {}", path)
+            } else {
+                anyhow::anyhow!("无法访问文件 {}: {}", path, e)
+            }
+        })?;
+    let metadata = file.metadata().await
+        .map_err(|e| anyhow::anyhow!("读取元数据失败 {}: {}", path, e))?;
+    if metadata.len() > 1_048_576 {
+        anyhow::bail!("文件超过 1MB 限制 ({} 字节)", metadata.len());
+    }
+    let mut raw_bytes = Vec::with_capacity(metadata.len() as usize);
+    file.read_to_end(&mut raw_bytes).await
+        .map_err(|e| anyhow::anyhow!("读取文件失败: {}", e))?;
+    let content = decode_bytes_to_string(&raw_bytes)
+        .map_err(|_| anyhow::anyhow!("文件非 UTF-8 文本(疑似二进制),无法对比: {}", path))?;
+    Ok((content, metadata.len()))
+}
+
 // ============================================================
 // patch_file 失败提示辅助 — old_text 不匹配时给 LLM 相近锚点
 // ============================================================