Add Chrome support to autowebcompat repro agent

agents/autowebcompat-repro/Dockerfile

@@ -28,19 +28,32 @@ ENV PATH="/opt/venv/bin:$PATH"
 
 FROM base AS agent
 
-# The Firefox DevTools MCP server is an npm package launched via `npx`, so the
-# agent image needs Node.js + npm (the python base ships neither). It also
-# needs the shared libraries Firefox requires to run headless; the Firefox
-# binary itself is downloaded at agent startup (a fresh Nightly per run) via
-# mozdownload/mozinstall, not baked in here.
+# The agent needs Node.js + npm to run the DevTools MCP servers (npm packages;
+# the python base ships neither) and the shared libraries the browsers require
+# to run headless. The browser binaries themselves are downloaded at agent
+# startup (a fresh build per run): Firefox via mozdownload/mozinstall and Chrome
+# for Testing via the Chrome-for-Testing JSON API (see browser.py).
 RUN apt-get update \
     && apt-get install -y --no-install-recommends \
+        # Node.js, to run the Firefox and Chrome DevTools MCP servers.
         nodejs npm \
+        # Shared by both browsers.
         ca-certificates \
-        libgtk-3-0 libdbus-glib-1-2 libx11-xcb1 libxtst6 libxt6 \
-        libasound2 libpci3 \
+        # Firefox runtime deps.
+        libgtk-3-0 libdbus-glib-1-2 libx11-xcb1 libxtst6 libxt6 libpci3 \
+        # Chrome runtime deps.
+        libnss3 libnspr4 libatk1.0-0t64 libatk-bridge2.0-0t64 libatspi2.0-0t64 \
+        libcups2t64 libdbus-1-3 libgbm1 libxcomposite1 libxdamage1 libxfixes3 \
+        libxrandr2 libxkbcommon0 libasound2t64 libpango-1.0-0 libcairo2 \
+        fonts-liberation \
     && rm -rf /var/lib/apt/lists/*
 
+# Install the DevTools MCP servers from the pinned package.json/lockfile
+COPY agents/autowebcompat-repro/package.json \
+     agents/autowebcompat-repro/package-lock.json \
+     /app/node/
+RUN cd /app/node && npm ci --omit=dev
+
 # hackbot.toml lives at the agent root (not inside the package), so copy it into
 # the working dir; the runtime discovers it there (cwd) at startup.
 COPY agents/autowebcompat-repro/hackbot.toml /app/hackbot.toml

agents/autowebcompat-repro/hackbot_agents/autowebcompat_repro/agent.py

@@ -28,9 +28,9 @@
 from hackbot_runtime.claude import Reporter
 from pydantic import BaseModel
 
-from .browser import FirefoxBrowsers
-from .config import BUGZILLA_READ_TOOLS, DEVTOOLS_TOOLS
-from .devtools_mcp import build_devtools_server
+from .browser import ChromeBrowsers, FirefoxBrowsers
+from .config import BUGZILLA_READ_TOOLS, CHROME_DEVTOOLS_TOOLS, DEVTOOLS_TOOLS
+from .mcp_servers import build_chrome_devtools_server, build_firefox_devtools_server
 from .result import (
     RESULT_SERVER_NAME,
     SUBMIT_RESULT_TOOL,
@@ -82,6 +82,7 @@ class AutowebcompatReproResult(BaseModel):
     plan_result: TestPlanResult
     reproductions: list[tuple[str, BugReproductionResult | ReproductionResult]]
     chrome_mask_fixed: bool | None
+    chrome_reproduced: bool | None
 
 
 @dataclass
@@ -317,6 +318,7 @@ def __init__(
         input_data: AutoWebcompatInput,
         bugzilla_mcp_server: McpServerConfig,
         screenshot_dir: Path,
+        chrome_path: Path,
     ):
         super().__init__(task_config, run_tracker)
         self.input_data = input_data
@@ -325,7 +327,7 @@ def __init__(
         )
         self.add_mcp_server(
             "firefox-devtools",
-            build_devtools_server(
+            build_firefox_devtools_server(
                 firefox_path=firefox_path,
                 headless=True,
                 enable_script=True,
@@ -334,7 +336,13 @@ def __init__(
             ),
             DEVTOOLS_TOOLS,
         )
-        if self.input_data.type == "bug_id" != None:
+
+        self.add_mcp_server(
+            "chrome-devtools",
+            build_chrome_devtools_server(chrome_path=chrome_path, headless=True),
+            CHROME_DEVTOOLS_TOOLS,
+        )
+        if self.input_data.type == "bug_id":
             self.add_mcp_server("bugzilla", bugzilla_mcp_server, BUGZILLA_READ_TOOLS)
 
     def subject(self) -> Any:
@@ -347,16 +355,30 @@ def system_prompt(self) -> str:
             .format(
                 task_details=f"""
 1. Identify the affected URL and the described broken behavior.
-2. Baseline: Navigate to the URL with the Firefox DevTools MCP and
-   try to reproduce the described broken behaviour.
-3. If the issue reproduces AND the breakage is visual in nature (incorrect
-   layout or rendering, not broken interaction), capture a screenshot showing
-   it: call `screenshot_page` with `saveTo` set to `{self.screenshot_path}`.
-   This writes the image to that file instead of returning it — do not capture
-   or paste the image data yourself. Then set `screenshot_path` in your result
-   to exactly `{self.screenshot_path}`. For non-visual issues, take no
-   screenshot and leave `screenshot_path` null.
-4. Submit your findings via `submit_result` (see "Reporting your result").
+2. Baseline: Navigate to the URL with the Firefox DevTools MCP
+   and try to reproduce the described broken behaviour.
+   - Reproduce against the actual reported site. If you cannot reach or
+     reproduce on that site — e.g. it is behind a login wall, blocked,
+     gated by a captcha, or down — do not substitute a reduced testcase,
+     minimal example, or other bug attachment as a stand-in reproduction.
+     Report `reproduced` as false with the appropriate `failure_reason`. A
+     testcase may inform your analysis, but reproducing on it does not count
+     as reproducing the reported site issue.
+3. Cross-check in Chrome: run the same steps in Chrome using the Chrome DevTools
+   MCP and report `chrome_reproduced`.
+   - A genuine web-compat issue reproduces in Firefox but not in Chrome. If the
+     behavior is identical in both, your steps may be wrong or this may not be a
+     web-compat issue; refine the steps and re-check before concluding.
+   - If the reported broken behaviour reproduces in both browsers, it is not a
+     Firefox web-compat issue: set `failure_reason` to `non_compat`.
+4. If the issue reproduces AND the breakage is visual in nature (incorrect
+   layout or rendering, not broken interaction), capture a screenshot in Firefox
+   showing it: call `screenshot_page` with `saveTo` set to
+   `{self.screenshot_path}`. This writes the image to that file instead of
+   returning it — do not capture or paste the image data yourself. Then set
+   `screenshot_path` in your result to exactly `{self.screenshot_path}`. For
+   non-visual issues, take no screenshot and leave `screenshot_path` null.
+5. Submit your findings via `submit_result` (see "Reporting your result").
 """
             )
         )
@@ -391,7 +413,7 @@ def __init__(
         self.steps = steps
         self.add_mcp_server(
             "firefox_devtools",
-            build_devtools_server(
+            build_firefox_devtools_server(
                 firefox_path=firefox_path,
                 headless=True,
                 enable_script=True,
@@ -437,7 +459,7 @@ def __init__(
         self.steps = steps
         self.add_mcp_server(
             "firefox_devtools",
-            build_devtools_server(
+            build_firefox_devtools_server(
                 firefox_path=firefox_path,
                 headless=True,
                 enable_script=True,
@@ -491,6 +513,7 @@ class InitialReproduction:
     steps: str
     summary: str
     screenshot_path: Path | None
+    chrome_reproduced: bool | None
 
 
 class ReproductionResults:
@@ -507,6 +530,19 @@ def __init__(self, publish_file: PublishFile, plan_result: TestPlanResult):
     def reproduced(self) -> bool:
         return self.initial_repro is not None
 
+    @property
+    def chrome_reproduced(self) -> bool | None:
+        if self.initial_repro is not None:
+            return self.initial_repro.chrome_reproduced
+
+        chrome_reproductions = [
+            result.chrome_reproduced
+            for result in self.results.values()
+            if isinstance(result, BugReproductionResult)
+            and result.chrome_reproduced is not None
+        ]
+        return any(chrome_reproductions) if chrome_reproductions else None
+
     @property
     def summary(self) -> str:
         return self.initial_repro.summary if self.initial_repro is not None else ""
@@ -555,7 +591,11 @@ def set_result(
             if self.initial_repro is not None:
                 raise ValueError("Got duplicate steps / summary")
             self.initial_repro = InitialReproduction(
-                channel, result.steps, result.summary, result.screenshot_path
+                channel,
+                result.steps,
+                result.summary,
+                result.screenshot_path,
+                result.chrome_reproduced,
             )
         elif isinstance(result, ChromeMaskResult):
             if self.chrome_mask_fixed is not None:
@@ -578,6 +618,7 @@ def into_result(self) -> AutowebcompatReproResult:
                 if isinstance(value, ReproductionResult)
             ],
             chrome_mask_fixed=self.chrome_mask_fixed,
+            chrome_reproduced=self.chrome_reproduced,
         )
 
 
@@ -594,6 +635,7 @@ async def run_autowebcompat_repro(
     :class:`AgentError` if the agent ends in an error.
     """
     firefox_browser = FirefoxBrowsers()
+    chrome_browser = ChromeBrowsers()
 
     test_plan_task = TestPlan(default_config, tracker, input_data, bugzilla_mcp_server)
     test_plan_result = await test_plan_task.run()
@@ -626,6 +668,7 @@ async def next_repro_task(
                 input_data,
                 bugzilla_mcp_server,
                 screenshots_dir,
+                chrome_browser.stable,
             )
         else:
             task = StepsReproduction(
@@ -647,6 +690,19 @@ async def next_repro_task(
     # Always try in nightly first
     await next_repro_task(FirefoxChannel.nightly)
 
+    # If the reported behavior reproduced in both Firefox and Chrome it is not a
+    # Firefox web-compat issue: stop early.
+    if repro_results.reproduced and repro_results.chrome_reproduced:
+        result = repro_results.into_result()
+        if result.failure_reason != "non_compat":
+            logger.warning(
+                "Issue reproduced in both Firefox and Chrome but failure_reason "
+                "was %r, should have been non_compat",
+                result.failure_reason,
+            )
+        result.failure_reason = "non_compat"
+        return result
+
     if not repro_results.reproduced and test_plan_result.affects_platforms == [
         "android"
     ]:

agents/autowebcompat-repro/hackbot_agents/autowebcompat_repro/browser.py

@@ -1,11 +1,16 @@
 import logging
+import os
 import platform
+import stat
+import sys
 import tempfile
+import zipfile
 from pathlib import Path
 from typing import Literal
 
 import mozdownload
 import mozinstall
+import requests
 
 logger = logging.getLogger("autowebcompat-repro")
 
@@ -72,3 +77,138 @@ def esr(self) -> Path:
         if self._esr is None:
             self._esr = install_firefox(channel="esr")
         return self._esr
+
+
+def chrome_platform() -> str:
+    """Chrome for Testing platform string for the current host."""
+    system = platform.system()
+    machine = platform.machine().lower()
+    if system == "Linux":
+        if machine not in {"x86_64", "amd64"}:
+            raise RuntimeError(
+                "Chrome for Testing has no linux build for "
+                f"{platform.machine()}; only x86_64/amd64 is supported. Run the "
+                "agent image as linux/amd64, e.g. DOCKER_DEFAULT_PLATFORM=linux/amd64."
+            )
+        return "linux64"
+    if system == "Darwin":
+        return "mac-arm64" if machine in {"arm64", "aarch64"} else "mac-x64"
+    if system == "Windows":
+        return "win64" if machine in {"x86_64", "amd64"} else "win32"
+    raise RuntimeError(f"Unsupported platform for Chrome for Testing: {system}")
+
+
+def resolve_chrome_download_url(channel: str, cft_platform: str) -> str:
+    """Look up the Chrome for Testing download URL for a channel + platform."""
+    versions_url = (
+        "https://googlechromelabs.github.io/chrome-for-testing/"
+        "last-known-good-versions-with-downloads.json"
+    )
+    response = requests.get(versions_url, timeout=120)
+    response.raise_for_status()
+    data = response.json()
+
+    entry = data["channels"][channel.capitalize()]
+    logger.info("Chrome for Testing %s: version %s", channel, entry["version"])
+
+    for download in entry["downloads"]["chrome"]:
+        if download["platform"] == cft_platform:
+            return download["url"]
+
+    raise RuntimeError(
+        f"no Chrome for Testing '{cft_platform}' download in {channel} channel"
+    )
+
+
+def chrome_binary_path(install_dir: Path, cft_platform: str) -> Path:
+    """Path to the Chrome for Testing binary within the unpacked archive.
+
+    Chrome for Testing uses a different executable name/layout per platform:
+    on macOS it is inside an `.app` bundle, on Windows it is `chrome.exe`,
+    and on Linux it is a `chrome`.
+    """
+    package = install_dir / f"chrome-{cft_platform}"
+    if cft_platform.startswith("mac"):
+        return (
+            package
+            / "Google Chrome for Testing.app"
+            / "Contents"
+            / "MacOS"
+            / "Google Chrome for Testing"
+        )
+    if cft_platform.startswith("win"):
+        return package / "chrome.exe"
+    return package / "chrome"
+
+
+def unzip(archive: Path, dest: Path) -> None:
+    """Extract a zip, preserving unix permission bits and recreating symlinks.
+
+    This keeps the Chrome binary executable and, on macOS, keeps the ``.app``
+    bundle's internal symlinks intact.
+
+    Adapted from wpt's tools/wpt/utils.py::unzip.
+    """
+    with zipfile.ZipFile(archive) as zip_data:
+        for info in zip_data.infolist():
+            # external_attr's two high bytes carry the unix st_mode, but only
+            # when the archive was created on a unix system (create_system == 3).
+            # A DOS/Windows-created archive, or extraction on Windows, carries no
+            # useful permission info, so fall back to a plain extract there.
+            if info.create_system == 0 or sys.platform == "win32":
+                zip_data.extract(info, path=dest)
+                continue
+
+            st_mode = info.external_attr >> 16
+            dst_path = os.path.join(dest, info.filename)
+            if stat.S_ISLNK(st_mode):
+                # Symlinks are stored as files whose contents are the target;
+                # recreate the link rather than extracting it as a file.
+                link_target = zip_data.read(info)
+                os.makedirs(os.path.dirname(dst_path), exist_ok=True)
+                if os.path.islink(dst_path):
+                    os.unlink(dst_path)
+                os.symlink(link_target, dst_path)
+            else:
+                zip_data.extract(info, path=dest)
+                # Preserve the permission bits (rwxrwxrwx) only, dropping the
+                # sticky/setuid/setgid bits.
+                os.chmod(dst_path, st_mode & 0o777)
+
+
+def install_chrome(channel: Literal["stable"] = "stable") -> Path:
+    """Download Chrome for Testing and return the browser binary path."""
+    cft_platform = chrome_platform()
+    install_dir = Path(tempfile.mkdtemp(prefix=f"chrome-{channel}-", dir=Path.home()))
+
+    url = resolve_chrome_download_url(channel, cft_platform)
+    archive = install_dir / f"chrome-{cft_platform}.zip"
+
+    logger.info("downloading Chrome for Testing from %s", url)
+    with requests.get(url, stream=True, timeout=120) as response:
+        response.raise_for_status()
+        with archive.open("wb") as out:
+            for chunk in response.iter_content(chunk_size=1 << 20):
+                if chunk:
+                    out.write(chunk)
+
+    unzip(archive, install_dir)
+    archive.unlink()
+
+    binary = chrome_binary_path(install_dir, cft_platform)
+    if not binary.exists():
+        raise RuntimeError(f"Chrome binary not found at {binary} after unpacking")
+
+    logger.info("installed Chrome at %s", binary)
+    return binary
+
+
+class ChromeBrowsers:
+    def __init__(self) -> None:
+        self._stable: Path | None = None
+
+    @property
+    def stable(self) -> Path:
+        if self._stable is None:
+            self._stable = install_chrome(channel="stable")
+        return self._stable