Add webcompat-triage agent

agents/autowebcompat-repro/Dockerfile

@@ -0,0 +1,64 @@
+FROM python:3.12 AS builder
+
+COPY --from=ghcr.io/astral-sh/uv:latest /uv /uvx /bin/
+
+ENV UV_PROJECT_ENVIRONMENT=/opt/venv
+
+WORKDIR /app
+
+# Install external deps without building workspace members.
+RUN --mount=type=cache,target=/root/.cache/uv \
+    --mount=type=bind,source=pyproject.toml,target=pyproject.toml \
+    --mount=type=bind,source=uv.lock,target=uv.lock \
+    --mount=type=bind,source=VERSION,target=VERSION \
+    uv sync --frozen --no-dev --no-install-workspace --package hackbot-agent-autowebcompat-repro
+
+RUN --mount=type=cache,target=/root/.cache/uv \
+    --mount=type=bind,target=/app,rw \
+    uv sync --locked --no-dev --no-editable --package hackbot-agent-autowebcompat-repro
+
+FROM python:3.12 AS base
+
+COPY --from=builder /opt/venv /opt/venv
+WORKDIR /app
+
+ENV PYTHONUNBUFFERED=1
+ENV PYTHONDONTWRITEBYTECODE=1
+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.
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends \
+        nodejs npm \
+        ca-certificates \
+        libgtk-3-0 libdbus-glib-1-2 libx11-xcb1 libxtst6 libxt6 \
+        libasound2 libpci3 \
+    && rm -rf /var/lib/apt/lists/*
+
+# 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
+
+RUN useradd --create-home --shell /bin/bash agent \
+    && mkdir -p /workspace \
+    && chown agent:agent /workspace
+
+USER agent
+
+CMD ["python", "-m", "hackbot_agents.autowebcompat_repro"]
+
+FROM base AS broker
+
+RUN useradd --create-home --shell /bin/bash broker
+
+USER broker
+
+EXPOSE 8765
+
+CMD ["python", "-m", "hackbot_agents.autowebcompat_repro.broker"]

agents/autowebcompat-repro/compose.yml

@@ -0,0 +1,31 @@
+services:
+  autowebcompat-repro-broker:
+    build:
+      context: ../..
+      dockerfile: agents/autowebcompat-repro/Dockerfile
+      target: broker
+    environment:
+      BUGZILLA_API_URL: ${BUGZILLA_API_URL}
+      BUGZILLA_API_KEY: ${BUGZILLA_API_KEY}
+    expose:
+      - "8765"
+
+  autowebcompat-repro-agent:
+    build:
+      context: ../..
+      dockerfile: agents/autowebcompat-repro/Dockerfile
+      target: agent
+    environment:
+      - RUN_ID
+      - BUG_DATA
+      - BUG_ID
+      - BUGZILLA_MCP_URL=http://autowebcompat-repro-broker:8765/mcp
+      - ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY:?error}
+      # No uploader locally: summary/logs/attachments are written under
+      # /artifacts/<run_id>, bind-mounted to the host's ~/hackbot/artifacts.
+      - ARTIFACTS_DIR=/artifacts
+    volumes:
+      - ${HOME}/hackbot/artifacts:/artifacts
+    depends_on:
+      autowebcompat-repro-broker:
+        condition: service_started

agents/autowebcompat-repro/hackbot.toml

@@ -0,0 +1,3 @@
+# autowebcompat-repro needs no platform prep: no [source] checkout, no [firefox] build.
+# Subject comes from the request (bug_data / bug_id); the DevTools MCP drives a
+# Firefox instance installed in the image.

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

@@ -0,0 +1,43 @@
+from hackbot_runtime import HackbotContext, run_async
+from pydantic_settings import BaseSettings, SettingsConfigDict
+
+from .agent import AutowebcompatReproResult, run_autowebcompat_repro
+from .firefox_install import install_firefox_nightly
+
+
+class AgentInputs(BaseSettings):
+    bugzilla_mcp_url: str
+    bug_data: str | None = None
+    bug_id: int | None = None
+    model: str | None = None
+    max_turns: int | None = None
+    effort: str | None = None
+
+    model_config = SettingsConfigDict(extra="ignore")
+
+
+async def main(ctx: HackbotContext) -> AutowebcompatReproResult:
+    inputs = AgentInputs()
+
+    # Provision a fresh Nightly at startup so each run reproduces against a
+    # current build; drive the binary the install reports back.
+    firefox_path = str(install_firefox_nightly())
+
+    return await run_autowebcompat_repro(
+        bugzilla_mcp_server={
+            "type": "http",
+            "url": inputs.bugzilla_mcp_url,
+        },
+        bug_data=inputs.bug_data,
+        bug_id=inputs.bug_id,
+        model=inputs.model,
+        max_turns=inputs.max_turns,
+        effort=inputs.effort,
+        firefox_path=firefox_path,
+        log=ctx.log_path,
+        verbose=True,
+    )
+
+
+if __name__ == "__main__":
+    run_async(main)

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

@@ -0,0 +1,155 @@
+"""Firefox web-compatibility reproduction agent.
+
+Drives an agent that reproduces a broken-site report in Firefox
+using the Firefox DevTools MCP. The bug is passed either inline as ``bug_data``
+text or a Bugzilla ``bug_id`` (read via Bugzilla broker).
+"""
+
+from __future__ import annotations
+
+import logging
+from pathlib import Path
+
+from claude_agent_sdk import (
+    ClaudeAgentOptions,
+    ClaudeSDKClient,
+    McpServerConfig,
+    ResultMessage,
+)
+from hackbot_runtime import AgentError, HackbotAgentResult
+from hackbot_runtime.claude import Reporter
+
+from .config import BUGZILLA_READ_TOOLS, DEVTOOLS_TOOLS
+from .devtools_mcp import build_devtools_server
+from .result import (
+    RESULT_SERVER_NAME,
+    SUBMIT_RESULT_TOOL,
+    ReproductionResult,
+    ResultCollector,
+    build_result_server,
+)
+
+HERE = Path(__file__).resolve().parent
+
+logger = logging.getLogger("autowebcompat-repro")
+
+
+class AutowebcompatReproResult(HackbotAgentResult):
+    result: ReproductionResult | None = None
+
+
+def load_system_prompt() -> str:
+    return (HERE / "prompts" / "system.md").read_text()
+
+
+def build_user_prompt(bug_data: str | None, bug_id: int | None) -> str:
+    if bug_data:
+        return (
+            "Here is the web-compatibility report to work on:\n\n"
+            f"{bug_data}\n\n"
+            "Follow your task procedure."
+        )
+    if bug_id is not None:
+        return (
+            f"The web-compatibility report to work on is Bugzilla bug {bug_id}. "
+            "Fetch it using the Bugzilla MCP tools, then follow your task procedure."
+        )
+    raise AgentError("neither bug_data nor bug_id was provided")
+
+
+async def run_autowebcompat_repro(
+    *,
+    bugzilla_mcp_server: McpServerConfig,
+    bug_data: str | None = None,
+    bug_id: int | None = None,
+    model: str | None = None,
+    max_turns: int | None = None,
+    effort: str | None = None,
+    firefox_path: str | None = None,
+    verbose: bool = False,
+    log: Path | None = None,
+) -> AutowebcompatReproResult:
+    """Reproduce a web-compat issue and return the agent's findings.
+
+    Returns a :class:`AutowebcompatReproResult` on success; raises
+    :class:`AgentError` if the agent ends in an error.
+    """
+    subject = bug_data if bug_data else f"bug {bug_id}"
+    preview = subject if len(subject) <= 200 else f"{subject[:200]}..."
+    logger.info("reproducing %s", preview)
+
+    devtools_server = build_devtools_server(
+        firefox_path=Path(firefox_path) if firefox_path else None,
+        headless=True,
+        enable_script=True,
+    )
+
+    # Structured-result MCP server (in-process): the agent calls submit_result
+    # once at the end, giving a predictable JSON result instead of free text.
+    result_collector = ResultCollector()
+    result_server = build_result_server(result_collector)
+
+    # Only wire up Bugzilla when there's a bug to fetch. With inline bug_data
+    # there's nothing to read, so the bugzilla MCP is not available
+    mcp_servers: dict[str, McpServerConfig] = {
+        "firefox-devtools": devtools_server,
+        RESULT_SERVER_NAME: result_server,
+    }
+    bugzilla_tools: list[str] = []
+    if bug_id is not None:
+        mcp_servers["bugzilla"] = bugzilla_mcp_server
+        bugzilla_tools = BUGZILLA_READ_TOOLS
+
+    system_prompt = load_system_prompt()
+
+    options = ClaudeAgentOptions(
+        system_prompt=system_prompt,
+        mcp_servers=mcp_servers,
+        permission_mode="bypassPermissions",
+        allowed_tools=[
+            "Read",
+            "Grep",
+            "Glob",
+            "Bash",
+            *bugzilla_tools,
+            *DEVTOOLS_TOOLS,
+            SUBMIT_RESULT_TOOL,
+        ],
+        model=model,
+        max_turns=max_turns,
+        **({"effort": effort} if effort else {}),
+        setting_sources=[],
+        # DevTools snapshots/screenshots of complex pages serialize to JSON that
+        # can exceed the SDK's default 1 MiB message buffer (the reader dies
+        # fatally if it does). Raise it well above that ceiling.
+        max_buffer_size=10 * 1024 * 1024,
+    )
+
+    user_prompt = build_user_prompt(bug_data, bug_id)
+
+    result_msg: ResultMessage | None = None
+    with Reporter(verbose=verbose, log_path=log) as reporter:
+        reporter.header(subject)
+        async with ClaudeSDKClient(options=options) as client:
+            await client.query(user_prompt)
+            async for msg in client.receive_response():
+                reporter.message(msg)
+                if isinstance(msg, ResultMessage):
+                    result_msg = msg
+
+    if result_msg is None:
+        raise AgentError(f"{subject}: agent produced no result message")
+    if result_msg.is_error:
+        raise AgentError(
+            f"{subject} investigation failed: {result_msg.result or result_msg.subtype}"
+        )
+    if result_collector.result is None:
+        raise AgentError(
+            f"{subject}: agent finished without submitting a result via submit_result"
+        )
+
+    return AutowebcompatReproResult(
+        result=result_collector.result,
+        num_turns=result_msg.num_turns,
+        total_cost_usd=result_msg.total_cost_usd,
+    )

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

@@ -0,0 +1,71 @@
+"""Bugzilla MCP broker.
+
+Sidecar container that holds the Bugzilla API key and serves the
+bugzilla MCP tools over HTTP. The agent process (in a sibling container
+in the same Cloud Run Job task) reaches us at `127.0.0.1:<port>/mcp`.
+The agent container itself binds no Bugzilla credentials.
+"""
+
+import logging
+from contextlib import asynccontextmanager
+
+import bugsy
+import uvicorn
+from agent_tools import bugzilla
+from agent_tools.bugzilla import BugzillaContext
+from agent_tools.claude_sdk import build_sdk_server
+from mcp.server.streamable_http_manager import StreamableHTTPSessionManager
+from pydantic_settings import BaseSettings, SettingsConfigDict
+from starlette.applications import Starlette
+from starlette.routing import Mount
+
+log = logging.getLogger("autowebcompat-repro-broker")
+
+
+class BrokerInputs(BaseSettings):
+    bugzilla_api_url: str
+    bugzilla_api_key: str
+    host: str = "0.0.0.0"
+    port: int = 8765
+
+    model_config = SettingsConfigDict(extra="ignore")
+
+
+def build_app(inputs: BrokerInputs) -> Starlette:
+    client = bugsy.Bugsy(
+        api_key=inputs.bugzilla_api_key, bugzilla_url=inputs.bugzilla_api_url
+    )
+    ctx = BugzillaContext(client=client)
+    sdk_config = build_sdk_server("bugzilla", ctx, bugzilla.TOOLS)
+    mcp_server = sdk_config["instance"]
+
+    manager = StreamableHTTPSessionManager(app=mcp_server, stateless=True)
+
+    @asynccontextmanager
+    async def lifespan(app):
+        async with manager.run():
+            log.info(
+                "bugzilla broker ready on %s:%d (read-only)",
+                inputs.host,
+                inputs.port,
+            )
+            yield
+
+    async def mcp_handler(scope, receive, send):
+        await manager.handle_request(scope, receive, send)
+
+    return Starlette(routes=[Mount("/mcp", app=mcp_handler)], lifespan=lifespan)
+
+
+def main() -> None:
+    logging.basicConfig(
+        level=logging.INFO,
+        format="%(asctime)s - %(name)s - %(levelname)s - %(message)s",
+    )
+    inputs = BrokerInputs()
+    app = build_app(inputs)
+    uvicorn.run(app, host=inputs.host, port=inputs.port, log_config=None)
+
+
+if __name__ == "__main__":
+    main()