dsh-eval-guard 0.3.4 → 0.3.5

package/README.md

@@ -32,7 +32,7 @@
 - 原 5 个比赛任务：坦克大战、仓储调度、圈地、炸弹人、夺旗。
 - 3 个长时单任务：Commit0 单库实现、SWE-Bench Pro 单任务、FeatureBench 单 feature。
 
-原 5 个任务直接放入打包好的 workspace.zip。后 3 个任务在安装时由 host 侧 setup 脚本按固定 commit clone 仓库、创建 .venv 并安装依赖；agent 运行时仍然禁网，使用工作区内的 .venv 跑测试。
+原 5 个任务直接放入打包好的 workspace.zip。后 3 个任务在安装时由 host 侧 setup 脚本按固定 commit clone 仓库、创建 .venv 并安装依赖；安装完成后会清理 `.git`、`setup.sh`、`remove.patch`、`spec.pdf.bz2` 等安装产物，工作区只保留任务本身素材。Commit0 的 `pytest.ini` 会去掉 coverage 参数，保证默认测试命令在沙箱内可运行。agent 运行时仍然禁网，使用工作区内的 .venv 跑测试；agent 可以自行 `git init` 或写自己的脚本。
 
 安装是后台任务：
 

package/lib/index.js

@@ -1998,6 +1998,21 @@ var TaskService = class {
 			});
 		}
 	}
+	/** Remove install-only artifacts so they are not part of the agent workspace. */
+	cleanupWorkspace(workspacePath) {
+		for (const name of [
+			".git",
+			".repo",
+			"setup.sh",
+			"remove.patch",
+			"spec.pdf.bz2"
+		]) try {
+			fs.rmSync(path.join(workspacePath, name), {
+				recursive: true,
+				force: true
+			});
+		} catch {}
+	}
 	/** Start a non-blocking install. The settings page polls the returned job. */
 	startInstall(workspaceId, taskId, options = {}) {
 		const definition = TASK_DEFINITIONS[taskId];
@@ -2049,6 +2064,7 @@ var TaskService = class {
 				cwd: checked.path,
 				env,
 				onSuccess: () => {
+					if (definition.needsSetup) this.cleanupWorkspace(checked.path);
 					const runId = this.newRunId();
 					const prompt = this.guard.promptFor(checked.path, this.materialZip(taskId), definition);
 					const run = {
@@ -2159,6 +2175,7 @@ var TaskService = class {
 			fs.copyFileSync(this.materialZip(taskId), path.join(checked.path, "workspace.zip"));
 			if (definition.needsSetup) this.extractMaterial(checked.path);
 			if (definition.prepare !== void 0) definition.prepare(checked.path);
+			if (definition.needsSetup) this.cleanupWorkspace(checked.path);
 		} catch (error) {
 			return {
 				ok: false,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-eval-guard",
-  "version": "0.3.4",
+  "version": "0.3.5",
   "type": "module",
   "description": "Evaluation guard for DSH: workspace-only filesystem, no network/subagents, wall-clock limits, and task installation for the meow-ai-arena tank task.",
   "license": "MIT",