fix(setup): honor sandbox image in BOM-prefixed configs (#5515)

* fix(setup): honor sandbox image in BOM-prefixed configs

* fix(setup): normalize CRLF and preserve captured pull arguments

README.md

@@ -403,6 +403,7 @@ such a checkout, use `bash ./scripts/<name>.sh ...`.
    # Recommended if using Docker/Container-based sandbox
    make setup-sandbox
    ```
+   Reads the configured sandbox image from UTF-8 `config.yaml`, with or without a leading BOM, using LF or CRLF line endings.
 
 4. **(Optional) Load sample memory data for local review**:
    ```bash

backend/tests/test_setup_sandbox.py

@@ -0,0 +1,45 @@
+"""Regression coverage for sandbox image selection before pulling."""
+
+import os
+import subprocess
+from pathlib import Path
+
+import pytest
+from support.shell import find_script_bash
+
+REPO_ROOT = Path(__file__).resolve().parents[2]
+BASH = find_script_bash()
+pytestmark = pytest.mark.skipif(BASH is None, reason="repo shell-script tests need Git Bash on Windows")
+
+
+@pytest.mark.parametrize("newline", ["\n", "\r\n"], ids=["lf", "crlf"])
+@pytest.mark.parametrize("encoding", ["utf-8", "utf-8-sig"])
+@pytest.mark.parametrize("configured", [True, False], ids=["configured", "commented"])
+def test_setup_sandbox_image_selection(tmp_path, encoding, configured, newline):
+    image = "example.invalid/custom-sandbox:review"
+    config = f"sandbox:\n  image: {image}\n" if configured else f"# sandbox:\n#   image: {image}\n"
+    (tmp_path / "config.yaml").write_bytes(config.replace("\n", newline).encode(encoding))
+    bin_dir = tmp_path / "bin"
+    bin_dir.mkdir()
+    capture = tmp_path / "docker-args.txt"
+    # Stub both container engines so the test never pulls a real image.
+    for name, script in {
+        "docker": '#!/usr/bin/env bash\nprintf "%s\\n" "$@" >> "$CAPTURE_DOCKER_ARGS"\n',
+        "container": "#!/usr/bin/env bash\nexit 0\n",
+    }.items():
+        path = bin_dir / name
+        path.write_text(script, encoding="utf-8")
+        path.chmod(0o755)
+    env = {**os.environ, "PATH": f"{bin_dir}{os.pathsep}{os.environ['PATH']}", "CAPTURE_DOCKER_ARGS": str(capture)}
+    result = subprocess.run([BASH, str(REPO_ROOT / "scripts/setup-sandbox.sh")], cwd=tmp_path, env=env, capture_output=True, text=True, check=True)
+    # Preserve carriage returns in arguments instead of normalizing them away.
+    args = capture.read_bytes().decode("utf-8").split("\n")[:-1]
+    assert args[0] == "pull"
+    assert len(args) == 2
+    if configured:
+        assert args[1] == image
+        assert f"Using configured image: {image}" in result.stdout
+        assert "Using default image:" not in result.stdout
+    else:
+        assert args[1] != image
+        assert f"Using default image: {args[1]}" in result.stdout

scripts/AGENTS.md

@@ -6,6 +6,8 @@ indentless lists are supported; nested option names and block-scalar text
 must not enable the browser extra. Keep the detector standard-library-only
 because it runs before dependency synchronization. Read UTF-8 config files
 with or without a leading BOM so the first section remains detectable.
+`setup-sandbox.sh` also strips the leading BOM and normalizes CRLF before selecting the image;
+keep its shell filter compatible with GNU and BSD sed.
 
 The root `PORT` value configures Docker's published nginx ingress only; local
 orchestration pins Next.js to `3000`. Runtime commands launch from the already

scripts/setup-sandbox.sh

@@ -12,8 +12,9 @@ echo ""
 IMAGE=""
 CONFIGURED=1
 if [ -f "config.yaml" ]; then
-    # Look for uncommented image: field under the sandbox section
-    IMAGE=$(grep -A 20 "^sandbox:" config.yaml 2>/dev/null | grep "^  image:" | awk '{print $2}' | head -1 || true)
+    # Strip a leading UTF-8 BOM and CRLF line endings before matching.
+    # Bash expands the bytes so this works with both GNU and BSD sed.
+    IMAGE=$(sed $'1s/^\xef\xbb\xbf//;s/\r$//' config.yaml 2>/dev/null | grep -A 20 "^sandbox:" | grep "^  image:" | awk '{print $2}' | head -1 || true)
 fi
 
 if [ -z "$IMAGE" ]; then