fix(init): skip binary repository reads

Author: betegon

packages/cli/src/lib/init/tools/read-files.ts

@@ -1,4 +1,5 @@
 import fs from "node:fs";
+import { isLikelyBinary } from "../../scan/index.js";
 import { MAX_FILE_BYTES } from "../constants.js";
 import type { ReadFilesPayload, ToolResult } from "../types.js";
 import { safePath } from "./shared.js";
@@ -42,23 +43,39 @@ async function readSingleFile(
     if (!stat.isFile()) {
       return null;
     }
-    if (stat.size <= maxBytes) {
-      return await fs.promises.readFile(absPath, "utf-8");
-    }
+    const buffer =
+      stat.size <= maxBytes
+        ? await fs.promises.readFile(absPath)
+        : await readFilePrefix(absPath, maxBytes);
 
-    const handle = await fs.promises.open(absPath, "r");
-    try {
-      const buffer = Buffer.alloc(maxBytes);
-      await handle.read(buffer, 0, maxBytes, 0);
-      return buffer.toString("utf-8");
-    } finally {
-      await handle.close();
+    // Repository listings can contain executables, archives, images, and
+    // other binary artifacts. Never decode those bytes into a prompt string:
+    // replacement characters and JSON escaping can turn a bounded file read
+    // into a much larger workflow payload with no useful project context.
+    if (isLikelyBinary(buffer)) {
+      return null;
     }
+
+    return buffer.toString("utf-8");
   } catch {
     return null;
   }
 }
 
+async function readFilePrefix(
+  absPath: string,
+  maxBytes: number
+): Promise<Buffer> {
+  const handle = await fs.promises.open(absPath, "r");
+  try {
+    const buffer = Buffer.alloc(maxBytes);
+    const { bytesRead } = await handle.read(buffer, 0, maxBytes, 0);
+    return buffer.subarray(0, bytesRead);
+  } finally {
+    await handle.close();
+  }
+}
+
 /**
  * Tool definition for batched file reads.
  */

packages/cli/src/lib/init/workflow-inputs.ts

@@ -1,5 +1,6 @@
 import fs from "node:fs";
 import path from "node:path";
+import { isLikelyBinary } from "../scan/index.js";
 import { MAX_FILE_BYTES } from "./constants.js";
 import { detectSentry } from "./tools/detect-sentry.js";
 import { listDir } from "./tools/list-dir.js";
@@ -131,7 +132,12 @@ export async function preReadCommonFiles(
       if (stat.size > MAX_FILE_BYTES) {
         continue;
       }
-      const content = await fs.promises.readFile(absPath, "utf-8");
+      const buffer = await fs.promises.readFile(absPath);
+      if (isLikelyBinary(buffer)) {
+        cache[filePath] = null;
+        continue;
+      }
+      const content = buffer.toString("utf-8");
       if (totalBytes + content.length <= MAX_PREREAD_TOTAL_BYTES) {
         cache[filePath] = content;
         totalBytes += content.length;

packages/cli/test/lib/init/tools/filesystem-tools.test.ts

@@ -103,6 +103,26 @@ describe("filesystem tools", () => {
     expect((existsResult.data as any).exists["missing.txt"]).toBe(false);
   });
 
+  test("does not decode binary files into workflow prompt content", async () => {
+    fs.writeFileSync(
+      path.join(testDir, "sentry-wizard"),
+      Buffer.from([0x7f, 0x45, 0x4c, 0x46, 0x02, 0x00, 0xff])
+    );
+
+    const result = await executeTool(
+      {
+        type: "tool",
+        operation: "read-files",
+        cwd: testDir,
+        params: { paths: ["sentry-wizard"] },
+      },
+      makeContext(testDir)
+    );
+
+    expect(result.ok).toBe(true);
+    expect((result.data as any).files["sentry-wizard"]).toBeNull();
+  });
+
   test("applies patchsets and injects auth tokens into env files", async () => {
     const result = await executeTool(
       {

packages/cli/test/lib/safe-read.test.ts

@@ -272,4 +272,18 @@ describe("workflow-inputs preReadCommonFiles FIFO safety", () => {
     expect(cache["package.json"]).toBe('{"name":"x"}');
     expect(cache["tsconfig.json"]).toBeNull();
   });
+
+  test("does not pre-read binary content disguised as a common config", async () => {
+    writeFileSync(
+      join(dir, "package.json"),
+      Buffer.from([0x7f, 0x45, 0x4c, 0x46, 0x02, 0x00, 0xff])
+    );
+
+    const listing: DirEntry[] = [
+      { name: "package.json", path: "package.json", type: "file" },
+    ];
+    const cache = await preReadCommonFiles(dir, listing);
+
+    expect(cache["package.json"]).toBeNull();
+  });
 });