server/tasks/medium_build.py

"""
Medium Task: Build Configuration

Scenario: A CI/CD pipeline is failing because a Python project is
missing critical dependencies in its requirements.txt. The agent
must read build error logs, identify the missing packages, and
patch the requirements file to fix the build.

Setup: Creates a Python project structure with a Flask app, test suite,
and a requirements.txt that is deliberately missing 'flask'.

Grader: Run the build script — exit 0 = pass, else fail.
"""

from __future__ import annotations

import os
import stat
import subprocess
import textwrap

WORKSPACE = "/tmp/sre_tasks/medium_build"
BUILD_SCRIPT = f"{WORKSPACE}/run_build.sh"
REQUIREMENTS = f"{WORKSPACE}/requirements.txt"
APP_FILE = f"{WORKSPACE}/app.py"
TEST_FILE = f"{WORKSPACE}/tests/test_app.py"
BUILD_LOG = f"{WORKSPACE}/build_output.log"

SYSTEM_PROMPT = textwrap.dedent("""\
    You are an SRE agent debugging a CI/CD pipeline failure.

    INCIDENT REPORT:
    - Alert: Build Pipeline FAILED — Stage "install & test" exited non-zero.
    - Impact: No new deployments can ship until the build is green.
    - Build workspace: /tmp/sre_tasks/medium_build/

    Your task:
    1. Read the build error log at /tmp/sre_tasks/medium_build/build_output.log
    2. Examine the project files to understand what's wrong.
    3. Fix the issue so that running `bash /tmp/sre_tasks/medium_build/run_build.sh`
       exits with code 0 (success).

    The project is a Python Flask web app with a test suite.
    The build script installs dependencies and runs tests.
""")


def setup() -> str:
    """
    Set up the medium_build task.

    Creates a Flask project with a deliberately broken requirements.txt
    (missing the 'flask' dependency). Pre-generates a realistic build
    failure log.

    Returns:
        The initial observation message.
    """
    # Clean workspace
    os.makedirs(f"{WORKSPACE}/tests", exist_ok=True)

    # Write the Flask app
    app_code = textwrap.dedent("""\
        from flask import Flask, jsonify

        app = Flask(__name__)

        @app.route("/api/status")
        def status():
            return jsonify({"status": "ok", "version": "1.2.0"})

        @app.route("/api/users")
        def users():
            return jsonify({"users": ["alice", "bob", "charlie"]})

        if __name__ == "__main__":
            app.run(host="0.0.0.0", port=5000)
    """)

    with open(APP_FILE, "w") as f:
        f.write(app_code)

    # Write the test suite
    test_code = textwrap.dedent("""\
        import pytest
        from app import app


        @pytest.fixture
        def client():
            app.config["TESTING"] = True
            with app.test_client() as client:
                yield client


        def test_status_endpoint(client):
            response = client.get("/api/status")
            assert response.status_code == 200
            data = response.get_json()
            assert data["status"] == "ok"


        def test_users_endpoint(client):
            response = client.get("/api/users")
            assert response.status_code == 200
            data = response.get_json()
            assert len(data["users"]) == 3
    """)

    with open(TEST_FILE, "w") as f:
        f.write(test_code)

    # Write BROKEN requirements.txt — missing flask!
    broken_requirements = textwrap.dedent("""\
        # Project dependencies
        pytest>=7.0
        requests>=2.28.0
    """)

    with open(REQUIREMENTS, "w") as f:
        f.write(broken_requirements)

    # Write the build script
    build_script = textwrap.dedent(f"""\
        #!/bin/bash
        set -e
        cd {WORKSPACE}
        pip install --quiet -r requirements.txt 2>&1
        cd {WORKSPACE} && python -m pytest tests/ -v 2>&1
        echo "BUILD SUCCESSFUL"
    """)

    with open(BUILD_SCRIPT, "w") as f:
        f.write(build_script)
    os.chmod(BUILD_SCRIPT, os.stat(BUILD_SCRIPT).st_mode | stat.S_IEXEC)

    # Generate a realistic build failure log
    build_log = textwrap.dedent("""\
        ===== CI/CD Build Pipeline =====
        Stage: install & test
        Timestamp: 2026-03-25 14:22:07 UTC

        [1/2] Installing dependencies from requirements.txt...
        Successfully installed pytest-7.4.0 requests-2.31.0

        [2/2] Running test suite...
        ============================= test session starts ==============================
        collected 0 items / 2 errors

        _____________________________ ERROR collecting tests/test_app.py ______________________________
        ImportError while importing test module '/workspace/tests/test_app.py'.
        Hint: make sure your test modules/packages have valid Python names.
        Traceback (most recent call last):
          File "/workspace/tests/test_app.py", line 2, in <module>
            from app import app
          File "/workspace/app.py", line 1, in <module>
            from flask import Flask, jsonify
        ModuleNotFoundError: No module named 'flask'
        =========================== short test summary info ============================
        ERROR tests/test_app.py - ModuleNotFoundError: No module named 'flask'
        ============================== 2 errors in 0.12s ==============================

        BUILD FAILED — exit code 1
    """)

    with open(BUILD_LOG, "w") as f:
        f.write(build_log)

    return (
        "ALERT: Build Pipeline FAILED — Stage 'install & test' exited non-zero.\n"
        "Build log: /tmp/sre_tasks/medium_build/build_output.log\n"
        "Build script: /tmp/sre_tasks/medium_build/run_build.sh\n"
        "Diagnose and fix the build failure."
    )


def grade() -> float:
    """
    Grade the medium_build task.

    Runs the build script. Returns 0.95 if it exits 0, else 0.05.
    """
    try:
        result = subprocess.run(
            ["bash", BUILD_SCRIPT],
            capture_output=True,
            text=True,
            timeout=60,
            cwd=WORKSPACE,
        )
        if result.returncode == 0:
            return 0.95
    except Exception:
        pass
    return 0.05


def cleanup() -> None:
    """No persistent processes to clean up for build tasks."""
    pass