release: 7 leaf sandbox tools — read/write/edit/grep/glob/todo/web_fetch (#585)

lib/agent/__tests__/buildAgentTools.test.ts

@@ -1,17 +1,35 @@
 import { describe, it, expect } from "vitest";
 import { buildAgentTools } from "@/lib/agent/buildAgentTools";
 
+const EXPECTED_TOOL_NAMES = [
+  "bash",
+  "read",
+  "write",
+  "edit",
+  "grep",
+  "glob",
+  "todo_write",
+  "web_fetch",
+] as const;
+
 describe("buildAgentTools", () => {
-  it("returns a tools record keyed by tool name", () => {
+  it("returns a tools record with all 8 leaf tools registered", () => {
     const tools = buildAgentTools();
-    expect(tools).toHaveProperty("bash");
-    expect(typeof tools.bash).toBe("object");
+    for (const name of EXPECTED_TOOL_NAMES) {
+      expect(tools).toHaveProperty(name);
+    }
   });
 
-  it("each tool has an inputSchema, description, and execute", () => {
-    const tools = buildAgentTools();
-    expect(tools.bash.inputSchema).toBeDefined();
-    expect(tools.bash.description).toBeDefined();
-    expect(typeof tools.bash.execute).toBe("function");
+  it("each tool exposes the AI SDK shape (description + inputSchema + execute)", () => {
+    const tools = buildAgentTools() as Record<
+      string,
+      { description?: unknown; inputSchema?: unknown; execute?: unknown }
+    >;
+    for (const name of EXPECTED_TOOL_NAMES) {
+      const t = tools[name]!;
+      expect(typeof t.description).toBe("string");
+      expect(t.inputSchema).toBeDefined();
+      expect(typeof t.execute).toBe("function");
+    }
   });
 });

lib/agent/buildAgentTools.ts

@@ -1,19 +1,38 @@
 import { bashTool } from "@/lib/agent/tools/bashTool";
+import { readFileTool } from "@/lib/agent/tools/readFileTool";
+import { writeFileTool } from "@/lib/agent/tools/writeFileTool";
+import { editFileTool } from "@/lib/agent/tools/editFileTool";
+import { grepTool } from "@/lib/agent/tools/grepTool";
+import { globTool } from "@/lib/agent/tools/globTool";
+import { todoWriteTool } from "@/lib/agent/tools/todoWriteTool";
+import { webFetchTool } from "@/lib/agent/tools/webFetchTool";
 
 /**
  * Factory for the full agent tool set passed into `streamText({ tools })`.
  * Each tool reads its sandbox handle + recoup creds from `experimental_context`
  * at execute time — the factory takes no arguments because the tools are
  * stateless modulo that context.
  *
- * Slim PR 4 exposes only `bash`. The remaining sandbox tools (`read`,
- * `write`, `grep`, `glob`, `todo`, `task`, `ask_user_question`, `skill`,
- * `fetch`) port in follow-up PRs and slot into this record one-by-one
- * without changing the factory signature.
+ * Currently ships 8 leaf tools:
+ *   - bash, read, write, edit, grep, glob (sandbox / file ops)
+ *   - todo_write (planning surface; stateless, echoes the list back)
+ *   - web_fetch (HTTP via curl inside the sandbox)
+ *
+ * Composite tools (`task` subagent, `ask_user_question` UI part,
+ * `skill` skill discovery) port in a follow-up PR — they require
+ * subagent context plumbing / UI rendering / skill discovery infra
+ * that isn't in api today.
  */
 export function buildAgentTools() {
   return {
-    bash: bashTool(),
+    bash: bashTool,
+    read: readFileTool,
+    write: writeFileTool,
+    edit: editFileTool,
+    grep: grepTool,
+    glob: globTool,
+    todo_write: todoWriteTool,
+    web_fetch: webFetchTool,
   };
 }
 

lib/agent/tools/__tests__/bashTool.test.ts

@@ -34,7 +34,7 @@ describe("bashTool.execute", () => {
     });
     vi.mocked(connectVercel).mockResolvedValue(sandbox as never);
 
-    const tool = bashTool();
+    const tool = bashTool;
     const result = await tool.execute!({ command: "ls" }, {
       experimental_context: baseContext,
     } as never);
@@ -64,7 +64,7 @@ describe("bashTool.execute", () => {
     });
     vi.mocked(connectVercel).mockResolvedValue(sandbox as never);
 
-    const tool = bashTool();
+    const tool = bashTool;
     const result = (await tool.execute!({ command: "find ." }, {
       experimental_context: baseContext,
     } as never)) as { truncated?: boolean };
@@ -83,7 +83,7 @@ describe("bashTool.execute", () => {
     });
     vi.mocked(connectVercel).mockResolvedValue(sandbox as never);
 
-    const tool = bashTool();
+    const tool = bashTool;
     await tool.execute!({ command: "ls", cwd: "apps/web" }, {
       experimental_context: baseContext,
     } as never);
@@ -107,7 +107,7 @@ describe("bashTool.execute", () => {
     });
     vi.mocked(connectVercel).mockResolvedValue(sandbox as never);
 
-    const tool = bashTool();
+    const tool = bashTool;
     await tool.execute!({ command: "curl example.com" }, {
       experimental_context: { ...baseContext, recoupOrgId: "org-uuid" },
     } as never);
@@ -121,7 +121,7 @@ describe("bashTool.execute", () => {
     });
     vi.mocked(connectVercel).mockResolvedValue(sandbox as never);
 
-    const tool = bashTool();
+    const tool = bashTool;
     const result = (await tool.execute!({ command: "npm run dev", detached: true }, {
       experimental_context: baseContext,
     } as never)) as { success: boolean; stdout: string };
@@ -134,7 +134,7 @@ describe("bashTool.execute", () => {
     const sandbox = makeSandbox({ execDetached: undefined });
     vi.mocked(connectVercel).mockResolvedValue(sandbox as never);
 
-    const tool = bashTool();
+    const tool = bashTool;
     const result = (await tool.execute!({ command: "npm run dev", detached: true }, {
       experimental_context: baseContext,
     } as never)) as { success: boolean; stderr: string };
@@ -148,7 +148,7 @@ describe("bashTool.execute", () => {
     });
     vi.mocked(connectVercel).mockResolvedValue(sandbox as never);
 
-    const tool = bashTool();
+    const tool = bashTool;
     await tool.execute!({ command: "npm run dev", detached: true }, {
       experimental_context: { ...baseContext, recoupOrgId: "org-uuid" },
     } as never);

lib/agent/tools/__tests__/editFileTool.test.ts

@@ -0,0 +1,86 @@
+import { describe, it, expect, vi, beforeEach } from "vitest";
+import { editFileTool } from "@/lib/agent/tools/editFileTool";
+import { connectVercel } from "@/lib/sandbox/vercel/connect/connectVercel";
+
+vi.mock("@/lib/sandbox/vercel/connect/connectVercel", () => ({
+  connectVercel: vi.fn(),
+}));
+
+const ctx = { sandbox: { state: { sandboxName: "x" }, workingDirectory: "/sandbox/mono" } };
+
+function makeSandbox(initialContent: string) {
+  let stored = initialContent;
+  return {
+    workingDirectory: "/sandbox/mono",
+    readFile: vi.fn(async () => stored),
+    writeFile: vi.fn(async (_path: string, content: string) => {
+      stored = content;
+    }),
+    getStored: () => stored,
+  };
+}
+
+beforeEach(() => vi.clearAllMocks());
+
+describe("editFileTool", () => {
+  it("replaces a unique oldString once and reports the startLine", async () => {
+    const sb = makeSandbox("line one\nold value\nline three");
+    vi.mocked(connectVercel).mockResolvedValue(sb as never);
+    const tool = editFileTool;
+    const result = (await tool.execute!(
+      { filePath: "a.txt", oldString: "old value", newString: "new value" },
+      { experimental_context: ctx } as never,
+    )) as { success: boolean; replacements: number; startLine: number };
+    expect(result.success).toBe(true);
+    expect(result.replacements).toBe(1);
+    expect(result.startLine).toBe(2);
+    expect(sb.getStored()).toBe("line one\nnew value\nline three");
+  });
+
+  it("rejects when oldString === newString (no-op)", async () => {
+    const sb = makeSandbox("anything");
+    vi.mocked(connectVercel).mockResolvedValue(sb as never);
+    const tool = editFileTool;
+    const result = (await tool.execute!({ filePath: "a.txt", oldString: "x", newString: "x" }, {
+      experimental_context: ctx,
+    } as never)) as { success: boolean; error: string };
+    expect(result.success).toBe(false);
+    expect(result.error).toMatch(/must be different/);
+  });
+
+  it("rejects when oldString is not in the file", async () => {
+    const sb = makeSandbox("hello world");
+    vi.mocked(connectVercel).mockResolvedValue(sb as never);
+    const tool = editFileTool;
+    const result = (await tool.execute!(
+      { filePath: "a.txt", oldString: "missing", newString: "other" },
+      { experimental_context: ctx } as never,
+    )) as { success: boolean; error: string };
+    expect(result.success).toBe(false);
+    expect(result.error).toMatch(/not found/);
+  });
+
+  it("rejects ambiguous edits (multiple matches without replaceAll)", async () => {
+    const sb = makeSandbox("foo\nfoo\nbar");
+    vi.mocked(connectVercel).mockResolvedValue(sb as never);
+    const tool = editFileTool;
+    const result = (await tool.execute!({ filePath: "a.txt", oldString: "foo", newString: "baz" }, {
+      experimental_context: ctx,
+    } as never)) as { success: boolean; error: string };
+    expect(result.success).toBe(false);
+    expect(result.error).toMatch(/2 times/);
+  });
+
+  it("replaces all occurrences when replaceAll:true", async () => {
+    const sb = makeSandbox("foo bar foo baz foo");
+    vi.mocked(connectVercel).mockResolvedValue(sb as never);
+    const tool = editFileTool;
+    const result = (await tool.execute!(
+      { filePath: "a.txt", oldString: "foo", newString: "qux", replaceAll: true },
+      { experimental_context: ctx } as never,
+    )) as { success: boolean; replacements: number };
+    expect(result.success).toBe(true);
+    expect(result.replacements).toBe(3);
+    expect(sb.getStored()).toBe("qux bar qux baz qux");
+  });
+});

lib/agent/tools/__tests__/globTool.test.ts

@@ -0,0 +1,97 @@
+import { describe, it, expect, vi, beforeEach } from "vitest";
+import { globTool } from "@/lib/agent/tools/globTool";
+import { connectVercel } from "@/lib/sandbox/vercel/connect/connectVercel";
+
+vi.mock("@/lib/sandbox/vercel/connect/connectVercel", () => ({
+  connectVercel: vi.fn(),
+}));
+
+const ctx = { sandbox: { state: { sandboxName: "x" }, workingDirectory: "/sandbox/mono" } };
+
+function makeSandbox(exec: ReturnType<typeof vi.fn>) {
+  return { workingDirectory: "/sandbox/mono", exec };
+}
+
+beforeEach(() => vi.clearAllMocks());
+
+describe("globTool", () => {
+  it("parses `mtime\\tsize\\tpath` output into structured file entries", async () => {
+    // Two files, newest first (sort already happens server-side in the command).
+    const sb = makeSandbox(
+      vi.fn().mockResolvedValue({
+        success: true,
+        exitCode: 0,
+        stdout:
+          "1700000000.0\t512\t/sandbox/mono/src/index.ts\n1699999000.5\t256\t/sandbox/mono/src/util.ts",
+        stderr: "",
+        truncated: false,
+      }),
+    );
+    vi.mocked(connectVercel).mockResolvedValue(sb as never);
+    const tool = globTool;
+    const result = (await tool.execute!({ pattern: "**/*.ts" }, {
+      experimental_context: ctx,
+    } as never)) as {
+      success: boolean;
+      count: number;
+      files: Array<{ path: string; size: number; modifiedAt: string }>;
+    };
+    expect(result.success).toBe(true);
+    expect(result.count).toBe(2);
+    expect(result.files[0]?.path).toBe("src/index.ts");
+    expect(result.files[0]?.size).toBe(512);
+    expect(typeof result.files[0]?.modifiedAt).toBe("string"); // ISO
+  });
+
+  it("emits a recursive find (no -maxdepth) for `**/*.ts`", async () => {
+    const sb = makeSandbox(
+      vi.fn().mockResolvedValue({
+        success: true,
+        exitCode: 0,
+        stdout: "",
+        stderr: "",
+        truncated: false,
+      }),
+    );
+    vi.mocked(connectVercel).mockResolvedValue(sb as never);
+    const tool = globTool;
+    await tool.execute!({ pattern: "**/*.ts" }, { experimental_context: ctx } as never);
+    const cmd = sb.exec.mock.calls[0]?.[0] as string;
+    expect(cmd).not.toContain("-maxdepth");
+  });
+
+  it("emits -maxdepth 1 for a bare `*.json` pattern (no recursion)", async () => {
+    const sb = makeSandbox(
+      vi.fn().mockResolvedValue({
+        success: true,
+        exitCode: 0,
+        stdout: "",
+        stderr: "",
+        truncated: false,
+      }),
+    );
+    vi.mocked(connectVercel).mockResolvedValue(sb as never);
+    const tool = globTool;
+    await tool.execute!({ pattern: "*.json" }, { experimental_context: ctx } as never);
+    expect(sb.exec.mock.calls[0]?.[0]).toMatch(/-maxdepth\s+1/);
+  });
+
+  it("returns success:false on non-1 exit codes", async () => {
+    const sb = makeSandbox(
+      vi.fn().mockResolvedValue({
+        success: false,
+        exitCode: 2,
+        stdout: "err",
+        stderr: "",
+        truncated: false,
+      }),
+    );
+    vi.mocked(connectVercel).mockResolvedValue(sb as never);
+    const tool = globTool;
+    const result = (await tool.execute!({ pattern: "**/*.ts" }, {
+      experimental_context: ctx,
+    } as never)) as { success: boolean; error: string };
+    expect(result.success).toBe(false);
+    expect(result.error).toMatch(/exit 2/);
+  });
+});