@electric-ax/agents-server-conformance-tests 0.1.8 → 0.1.10

package/dist/index.cjs

@@ -2904,25 +2904,51 @@ function runElectricAgentsConformanceTests(config) {
 			const block = result.content[0];
 			return block?.type === `text` && block.text ? block.text : ``;
 		}
+		async function makeSandbox(workingDirectory) {
+			const { unrestrictedSandbox } = await import(`../../agents-runtime/src/sandbox/unrestricted`);
+			return unrestrictedSandbox({ workingDirectory });
+		}
 		(0, vitest.test)(`bash tool captures stdout and stderr`, async () => {
 			const { createBashTool } = await import(`../../agents-runtime/src/tools`);
-			const tool = createBashTool(`/tmp`);
-			const result = await tool.execute(`test-tc`, { command: `echo "hello" && echo "error" >&2` });
-			(0, vitest.expect)(firstText(result)).toContain(`hello`);
-			(0, vitest.expect)(firstText(result)).toContain(`error`);
-			(0, vitest.expect)(result.details.exitCode).toBe(0);
+			const sandbox = await makeSandbox(`/tmp`);
+			try {
+				const tool = createBashTool(sandbox);
+				const result = await tool.execute(`test-tc`, { command: `echo "hello" && echo "error" >&2` });
+				(0, vitest.expect)(firstText(result)).toContain(`hello`);
+				(0, vitest.expect)(firstText(result)).toContain(`error`);
+				(0, vitest.expect)(result.details.exitCode).toBe(0);
+			} finally {
+				await sandbox.dispose();
+			}
 		});
 		(0, vitest.test)(`bash tool enforces timeout`, async () => {
 			const { createBashTool } = await import(`../../agents-runtime/src/tools`);
-			const tool = createBashTool(`/tmp`);
-			const result = await tool.execute(`test-tc`, { command: `sleep 60` });
-			(0, vitest.expect)(result.details.timedOut).toBe(true);
+			const sandbox = await makeSandbox(`/tmp`);
+			try {
+				const tool = createBashTool(sandbox);
+				const result = await tool.execute(`test-tc`, { command: `sleep 60` });
+				(0, vitest.expect)(result.details.timedOut).toBe(true);
+			} finally {
+				await sandbox.dispose();
+			}
 		}, 35e3);
 		(0, vitest.test)(`read_file rejects paths outside working directory`, async () => {
 			const { createReadFileTool } = await import(`../../agents-runtime/src/tools`);
-			const tool = createReadFileTool(`/tmp/test-workdir`);
-			const result = await tool.execute(`test-tc`, { path: `../../etc/passwd` });
-			(0, vitest.expect)(firstText(result)).toContain(`outside the working directory`);
+			const fs = await import(`node:fs/promises`);
+			const dir = `/tmp/test-workdir-${Date.now()}`;
+			await fs.mkdir(dir, { recursive: true });
+			const sandbox = await makeSandbox(dir);
+			try {
+				const tool = createReadFileTool(sandbox);
+				const result = await tool.execute(`test-tc`, { path: `../../etc/passwd` });
+				(0, vitest.expect)(firstText(result)).toContain(`outside the working directory`);
+			} finally {
+				await sandbox.dispose();
+				await fs.rm(dir, {
+					recursive: true,
+					force: true
+				});
+			}
 		});
 		(0, vitest.test)(`read_file rejects binary files`, async () => {
 			const { createReadFileTool } = await import(`../../agents-runtime/src/tools`);
@@ -2937,10 +2963,15 @@ function runElectricAgentsConformanceTests(config) {
 				2,
 				255
 			]));
-			const tool = createReadFileTool(dir);
-			const result = await tool.execute(`test-tc`, { path: `test.bin` });
-			(0, vitest.expect)(firstText(result)).toContain(`binary file`);
-			await fs.rm(dir, { recursive: true });
+			const sandbox = await makeSandbox(dir);
+			try {
+				const tool = createReadFileTool(sandbox);
+				const result = await tool.execute(`test-tc`, { path: `test.bin` });
+				(0, vitest.expect)(firstText(result)).toContain(`binary file`);
+			} finally {
+				await sandbox.dispose();
+				await fs.rm(dir, { recursive: true });
+			}
 		});
 		(0, vitest.test)(`read_file rejects oversized files`, async () => {
 			const { createReadFileTool } = await import(`../../agents-runtime/src/tools`);
@@ -2950,10 +2981,15 @@ function runElectricAgentsConformanceTests(config) {
 			await fs.mkdir(dir, { recursive: true });
 			const bigPath = path.join(dir, `big.txt`);
 			await fs.writeFile(bigPath, `x`.repeat(600 * 1024));
-			const tool = createReadFileTool(dir);
-			const result = await tool.execute(`test-tc`, { path: `big.txt` });
-			(0, vitest.expect)(firstText(result)).toContain(`too large`);
-			await fs.rm(dir, { recursive: true });
+			const sandbox = await makeSandbox(dir);
+			try {
+				const tool = createReadFileTool(sandbox);
+				const result = await tool.execute(`test-tc`, { path: `big.txt` });
+				(0, vitest.expect)(firstText(result)).toContain(`too large`);
+			} finally {
+				await sandbox.dispose();
+				await fs.rm(dir, { recursive: true });
+			}
 		});
 		(0, vitest.test)(`web_search tool has correct interface`, async () => {
 			const { braveSearchTool } = await import(`../../agents-runtime/src/tools`);
@@ -2961,9 +2997,15 @@ function runElectricAgentsConformanceTests(config) {
 			(0, vitest.expect)(typeof braveSearchTool.execute).toBe(`function`);
 		});
 		(0, vitest.test)(`fetch_url tool has correct interface`, async () => {
-			const { fetchUrlTool } = await import(`../../agents-runtime/src/tools`);
-			(0, vitest.expect)(fetchUrlTool.name).toBe(`fetch_url`);
-			(0, vitest.expect)(typeof fetchUrlTool.execute).toBe(`function`);
+			const { createFetchUrlTool } = await import(`../../agents-runtime/src/tools`);
+			const sandbox = await makeSandbox(`/tmp`);
+			try {
+				const tool = createFetchUrlTool(sandbox);
+				(0, vitest.expect)(tool.name).toBe(`fetch_url`);
+				(0, vitest.expect)(typeof tool.execute).toBe(`function`);
+			} finally {
+				await sandbox.dispose();
+			}
 		});
 	});
 	(0, vitest.describe)(`Electric Agents Send — State Protocol Format`, () => {

package/dist/index.js

@@ -2880,25 +2880,51 @@ function runElectricAgentsConformanceTests(config) {
 			const block = result.content[0];
 			return block?.type === `text` && block.text ? block.text : ``;
 		}
+		async function makeSandbox(workingDirectory) {
+			const { unrestrictedSandbox } = await import(`../../agents-runtime/src/sandbox/unrestricted`);
+			return unrestrictedSandbox({ workingDirectory });
+		}
 		test(`bash tool captures stdout and stderr`, async () => {
 			const { createBashTool } = await import(`../../agents-runtime/src/tools`);
-			const tool = createBashTool(`/tmp`);
-			const result = await tool.execute(`test-tc`, { command: `echo "hello" && echo "error" >&2` });
-			expect(firstText(result)).toContain(`hello`);
-			expect(firstText(result)).toContain(`error`);
-			expect(result.details.exitCode).toBe(0);
+			const sandbox = await makeSandbox(`/tmp`);
+			try {
+				const tool = createBashTool(sandbox);
+				const result = await tool.execute(`test-tc`, { command: `echo "hello" && echo "error" >&2` });
+				expect(firstText(result)).toContain(`hello`);
+				expect(firstText(result)).toContain(`error`);
+				expect(result.details.exitCode).toBe(0);
+			} finally {
+				await sandbox.dispose();
+			}
 		});
 		test(`bash tool enforces timeout`, async () => {
 			const { createBashTool } = await import(`../../agents-runtime/src/tools`);
-			const tool = createBashTool(`/tmp`);
-			const result = await tool.execute(`test-tc`, { command: `sleep 60` });
-			expect(result.details.timedOut).toBe(true);
+			const sandbox = await makeSandbox(`/tmp`);
+			try {
+				const tool = createBashTool(sandbox);
+				const result = await tool.execute(`test-tc`, { command: `sleep 60` });
+				expect(result.details.timedOut).toBe(true);
+			} finally {
+				await sandbox.dispose();
+			}
 		}, 35e3);
 		test(`read_file rejects paths outside working directory`, async () => {
 			const { createReadFileTool } = await import(`../../agents-runtime/src/tools`);
-			const tool = createReadFileTool(`/tmp/test-workdir`);
-			const result = await tool.execute(`test-tc`, { path: `../../etc/passwd` });
-			expect(firstText(result)).toContain(`outside the working directory`);
+			const fs = await import(`node:fs/promises`);
+			const dir = `/tmp/test-workdir-${Date.now()}`;
+			await fs.mkdir(dir, { recursive: true });
+			const sandbox = await makeSandbox(dir);
+			try {
+				const tool = createReadFileTool(sandbox);
+				const result = await tool.execute(`test-tc`, { path: `../../etc/passwd` });
+				expect(firstText(result)).toContain(`outside the working directory`);
+			} finally {
+				await sandbox.dispose();
+				await fs.rm(dir, {
+					recursive: true,
+					force: true
+				});
+			}
 		});
 		test(`read_file rejects binary files`, async () => {
 			const { createReadFileTool } = await import(`../../agents-runtime/src/tools`);
@@ -2913,10 +2939,15 @@ function runElectricAgentsConformanceTests(config) {
 				2,
 				255
 			]));
-			const tool = createReadFileTool(dir);
-			const result = await tool.execute(`test-tc`, { path: `test.bin` });
-			expect(firstText(result)).toContain(`binary file`);
-			await fs.rm(dir, { recursive: true });
+			const sandbox = await makeSandbox(dir);
+			try {
+				const tool = createReadFileTool(sandbox);
+				const result = await tool.execute(`test-tc`, { path: `test.bin` });
+				expect(firstText(result)).toContain(`binary file`);
+			} finally {
+				await sandbox.dispose();
+				await fs.rm(dir, { recursive: true });
+			}
 		});
 		test(`read_file rejects oversized files`, async () => {
 			const { createReadFileTool } = await import(`../../agents-runtime/src/tools`);
@@ -2926,10 +2957,15 @@ function runElectricAgentsConformanceTests(config) {
 			await fs.mkdir(dir, { recursive: true });
 			const bigPath = path.join(dir, `big.txt`);
 			await fs.writeFile(bigPath, `x`.repeat(600 * 1024));
-			const tool = createReadFileTool(dir);
-			const result = await tool.execute(`test-tc`, { path: `big.txt` });
-			expect(firstText(result)).toContain(`too large`);
-			await fs.rm(dir, { recursive: true });
+			const sandbox = await makeSandbox(dir);
+			try {
+				const tool = createReadFileTool(sandbox);
+				const result = await tool.execute(`test-tc`, { path: `big.txt` });
+				expect(firstText(result)).toContain(`too large`);
+			} finally {
+				await sandbox.dispose();
+				await fs.rm(dir, { recursive: true });
+			}
 		});
 		test(`web_search tool has correct interface`, async () => {
 			const { braveSearchTool } = await import(`../../agents-runtime/src/tools`);
@@ -2937,9 +2973,15 @@ function runElectricAgentsConformanceTests(config) {
 			expect(typeof braveSearchTool.execute).toBe(`function`);
 		});
 		test(`fetch_url tool has correct interface`, async () => {
-			const { fetchUrlTool } = await import(`../../agents-runtime/src/tools`);
-			expect(fetchUrlTool.name).toBe(`fetch_url`);
-			expect(typeof fetchUrlTool.execute).toBe(`function`);
+			const { createFetchUrlTool } = await import(`../../agents-runtime/src/tools`);
+			const sandbox = await makeSandbox(`/tmp`);
+			try {
+				const tool = createFetchUrlTool(sandbox);
+				expect(tool.name).toBe(`fetch_url`);
+				expect(typeof tool.execute).toBe(`function`);
+			} finally {
+				await sandbox.dispose();
+			}
 		});
 	});
 	describe(`Electric Agents Send — State Protocol Format`, () => {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@electric-ax/agents-server-conformance-tests",
-  "version": "0.1.8",
+  "version": "0.1.10",
   "description": "Conformance test suite for Electric Agents server implementations",
   "author": "Durable Stream contributors",
   "license": "Apache-2.0",
@@ -34,8 +34,8 @@
     "dist"
   ],
   "dependencies": {
-    "@durable-streams/client": "https://pkg.pr.new/durable-streams/durable-streams/@durable-streams/client@5d5c217",
-    "@electric-sql/client": "^1.5.19",
+    "@durable-streams/client": "^0.2.6",
+    "@electric-sql/client": "^1.5.20",
     "fast-check": "^4.6.0",
     "vitest": "^4.1.0"
   },
@@ -43,7 +43,7 @@
     "@mariozechner/pi-agent-core": "^0.70.2",
     "tsdown": "^0.9.0",
     "typescript": "^5.0.0",
-    "@electric-ax/agents-server": "0.4.11"
+    "@electric-ax/agents-server": "0.4.15"
   },
   "engines": {
     "node": ">=18.0.0"