tasks.py

"""Task definitions for LogSentinel v2."""

from models import DifficultyLevel, ScenarioConfig, TaskDefinition

TASKS: dict = {
    # ------------------------------------------------------------------
    # Legacy tasks (backward-compatible)
    # ------------------------------------------------------------------
    "log_classification": TaskDefinition(
        name="log_classification",
        description=(
            "Classify 10 log entries by severity and type from a single source. "
            "Each log should be classified as: normal, warning, error, critical, or security. "
            "Submit a final report when done."
        ),
        difficulty="easy",
        max_steps=15,
        num_logs=10,
        num_sources=1,
        num_incidents=0,
    ),
    "incident_detection": TaskDefinition(
        name="incident_detection",
        description=(
            "Analyze 20 log entries from 3 different sources. Classify each log, "
            "detect 2 incidents by correlating related logs across sources, "
            "and assign correct severity (P1-P4) to each incident. "
            "Submit a structured incident report when done."
        ),
        difficulty="medium",
        max_steps=25,
        num_logs=20,
        num_sources=3,
        num_incidents=2,
    ),
    "full_triage": TaskDefinition(
        name="full_triage",
        description=(
            "Full incident triage with 30+ log entries from 5 sources including noise. "
            "Classify all logs, detect 3 incidents (one is a subtle security issue), "
            "correlate logs across sources, assign severity, recommend remediation "
            "for each incident, and submit a comprehensive incident report. "
            "Includes red herrings and ambiguous logs."
        ),
        difficulty="hard",
        max_steps=35,
        num_logs=34,
        num_sources=5,
        num_incidents=3,
    ),
    # ------------------------------------------------------------------
    # Phase-2 Multi-Agent SOC War-Room tasks
    # ------------------------------------------------------------------
    "soc_warroom_easy": TaskDefinition(
        name="soc_warroom_easy",
        description=(
            "Multi-agent SOC scenario: detect and triage 1 incident across 4 roles "
            "(incident_commander, app_sre, db_sre, security_analyst). "
            "Each role sees only role-relevant logs. Use handoffs and shared board to collaborate. "
            "Episode phases: detect → triage → mitigate → verify → final_report."
        ),
        difficulty="easy",
        max_steps=30,
        num_logs=20,
        num_sources=4,
        num_incidents=1,
        multi_agent=True,
        scenario_config=ScenarioConfig(
            num_incidents=1,
            difficulty=DifficultyLevel.EASY,
            attack_subtlety=0.2,
            observability_quality=0.95,
            confounding_noise_ratio=0.25,
            seed=100,
        ),
    ),
    "soc_warroom_medium": TaskDefinition(
        name="soc_warroom_medium",
        description=(
            "Multi-agent SOC scenario: detect and triage 2 overlapping incidents. "
            "Partial observability, telemetry delays, and confounding noise. "
            "Roles must negotiate severity and share evidence via handoffs."
        ),
        difficulty="medium",
        max_steps=45,
        num_logs=35,
        num_sources=5,
        num_incidents=2,
        multi_agent=True,
        scenario_config=ScenarioConfig(
            num_incidents=2,
            difficulty=DifficultyLevel.MEDIUM,
            attack_subtlety=0.5,
            observability_quality=0.85,
            confounding_noise_ratio=0.35,
            seed=200,
        ),
    ),
    "soc_warroom_hard": TaskDefinition(
        name="soc_warroom_hard",
        description=(
            "Multi-agent SOC scenario: 3 simultaneous incidents including a subtle APT. "
            "High noise, low observability quality, telemetry delays. "
            "Requires careful evidence collection, cross-role handoffs, and coordinated mitigation."
        ),
        difficulty="hard",
        max_steps=60,
        num_logs=50,
        num_sources=5,
        num_incidents=3,
        multi_agent=True,
        scenario_config=ScenarioConfig(
            num_incidents=3,
            difficulty=DifficultyLevel.HARD,
            attack_subtlety=0.8,
            observability_quality=0.70,
            confounding_noise_ratio=0.50,
            seed=300,
        ),
    ),
    "adaptive_curriculum": TaskDefinition(
        name="adaptive_curriculum",
        description=(
            "Adaptive difficulty task that auto-adjusts based on rolling success rate. "
            "Starts at easy, promotes to medium/hard when avg success > 70%, "
            "demotes when avg success < 35%."
        ),
        difficulty="adaptive",
        max_steps=45,
        num_logs=35,
        num_sources=5,
        num_incidents=2,
        multi_agent=True,
        scenario_config=ScenarioConfig(
            num_incidents=2,
            difficulty=DifficultyLevel.EASY,
            seed=None,  # randomised per episode
        ),
    ),
}

TASK_LIST = list(TASKS.keys())