feat: implement AI Director & Narrative Voting System (Phase 9)

Add complete AI Director system that transforms the survival simulation
into a user-driven interactive story with audience voting.

Backend:
- Add DirectorService for LLM-powered plot generation with fallback templates
- Add VoteManager for dual-channel voting (Twitch + Unity)
- Integrate 4-phase game loop: Simulation → Narrative → Voting → Resolution
- Add vote command parsing (!1, !2, !A, !B) in Twitch service
- Add type-safe LLM output handling with _coerce_int() helper
- Normalize voter IDs for case-insensitive duplicate prevention

Unity Client:
- Add NarrativeUI for cinematic event cards and voting progress bars
- Add 7 new event types and data models for director/voting events
- Add delayed subscription coroutine for NetworkManager timing
- Sync client timer with server's remaining_seconds to prevent drift

Documentation:
- Update README.md with AI Director features, voting commands, and event types

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This commit is contained in:
empty
2026-01-02 03:37:41 +08:00
parent 93fed8b9ca
commit 8915a4b074
10 changed files with 2048 additions and 3 deletions

View File

@@ -0,0 +1,562 @@
"""
AI Director Service - Narrative Control Module (Phase 9).
The Director acts as the Dungeon Master for the survival drama,
generating dramatic plot points and resolving audience votes.
"""
from __future__ import annotations
import json
import logging
import os
import random
import time
import uuid
from dataclasses import dataclass, field
from enum import Enum
from typing import Any
logger = logging.getLogger(__name__)
class GameMode(str, Enum):
"""Game engine operating modes."""
SIMULATION = "simulation" # Normal agent behavior
NARRATIVE = "narrative" # Director presents plot point
VOTING = "voting" # Audience voting window
RESOLUTION = "resolution" # Applying vote consequences
@dataclass(frozen=True)
class PlotChoice:
"""A choice option in a plot point."""
choice_id: str
text: str
effects: dict[str, Any] = field(default_factory=dict)
@dataclass
class PlotPoint:
"""A narrative event generated by the Director."""
plot_id: str
title: str
description: str
choices: list[PlotChoice]
ttl_seconds: int = 60
created_at: float = field(default_factory=time.time)
def to_dict(self) -> dict[str, Any]:
"""Convert to dictionary for broadcasting."""
return {
"plot_id": self.plot_id,
"title": self.title,
"description": self.description,
"choices": [
{"choice_id": c.choice_id, "text": c.text}
for c in self.choices
],
"ttl_seconds": self.ttl_seconds,
}
@dataclass
class ResolutionResult:
"""Result of resolving a plot point vote."""
plot_id: str
choice_id: str
message: str
effects: dict[str, Any] = field(default_factory=dict)
def to_dict(self) -> dict[str, Any]:
"""Convert to dictionary for broadcasting."""
return {
"plot_id": self.plot_id,
"choice_id": self.choice_id,
"message": self.message,
"effects_json": json.dumps(self.effects),
}
# Fallback templates when LLM is unavailable
FALLBACK_PLOT_TEMPLATES = [
{
"title": "Mysterious Footprints",
"description": "Strange footprints appear on the beach overnight. Someone - or something - has been watching the camp.",
"choices": [
PlotChoice("investigate", "Follow the tracks into the forest", {"risk": "medium", "reward": "discovery"}),
PlotChoice("fortify", "Strengthen camp defenses and wait", {"safety": "high", "mood_delta": -5}),
],
},
{
"title": "Supply Shortage",
"description": "The food stores are running dangerously low. Tension builds among the survivors.",
"choices": [
PlotChoice("ration", "Implement strict rationing for everyone", {"mood_delta": -10, "food_save": 2}),
PlotChoice("hunt", "Send a group on a risky hunting expedition", {"risk": "high", "food_gain": 3}),
],
},
{
"title": "Storm Warning",
"description": "Dark clouds gather on the horizon. A massive storm approaches the island.",
"choices": [
PlotChoice("shelter", "Everyone take shelter immediately", {"safety": "high", "mood_delta": 5}),
PlotChoice("salvage", "Quickly gather supplies before the storm hits", {"risk": "medium", "resource_gain": 2}),
],
},
{
"title": "Trust Crisis",
"description": "Accusations fly as valuable supplies go missing from the camp.",
"choices": [
PlotChoice("accuse", "Hold a trial to find the culprit", {"drama": "high", "relationship_delta": -5}),
PlotChoice("forgive", "Call for unity and move on together", {"mood_delta": 3, "trust": "restored"}),
],
},
{
"title": "Rescue Signal",
"description": "A faint light flickers on the distant horizon. Could it be a ship?",
"choices": [
PlotChoice("signal", "Build a massive signal fire on the beach", {"energy_delta": -15, "hope": "high"}),
PlotChoice("wait", "Wait and observe - it could be dangerous", {"safety": "medium", "mood_delta": -3}),
],
},
]
class DirectorService:
"""
AI Director for generating and resolving narrative events.
Uses LLM to create dramatic plot points based on world state.
"""
def __init__(self, llm_service=None) -> None:
"""
Initialize the Director service.
Args:
llm_service: Optional LLMService instance. If None, uses global instance.
"""
self._llm_service = llm_service
self._rng = random.Random()
self._current_plot: PlotPoint | None = None
self._plot_history: list[str] = [] # Recent plot titles to avoid repetition
@property
def llm(self):
"""Lazy-load LLM service to avoid circular imports."""
if self._llm_service is None:
from .llm import llm_service
self._llm_service = llm_service
return self._llm_service
@property
def current_plot(self) -> PlotPoint | None:
"""Get the current active plot point."""
return self._current_plot
def clear_current_plot(self) -> None:
"""Clear the current plot after resolution."""
if self._current_plot:
self._plot_history.append(self._current_plot.title)
# Keep only last 5 titles to avoid repetition
self._plot_history = self._plot_history[-5:]
self._current_plot = None
async def generate_plot_point(self, world_state: dict[str, Any]) -> PlotPoint:
"""
Generate a dramatic plot point based on current world state.
Args:
world_state: Dictionary containing:
- day: Current game day
- weather: Current weather condition
- time_of_day: dawn/day/dusk/night
- alive_agents: List of alive agent summaries
- recent_events: List of recent event descriptions
- tension_level: low/medium/high (derived from deaths, resources, etc.)
Returns:
PlotPoint with title, description, and 2 choices
"""
# Extract context
day = world_state.get("day", 1)
weather = world_state.get("weather", "Sunny")
alive_count = len(world_state.get("alive_agents", []))
recent_events = world_state.get("recent_events", [])
tension_level = world_state.get("tension_level", "medium")
mood_avg = world_state.get("mood_avg", 50)
# Build context summary
agents_summary = ", ".join([
f"{a.get('name', 'Unknown')} (HP:{a.get('hp', 0)})"
for a in world_state.get("alive_agents", [])[:5]
]) or "No agents alive"
events_summary = "; ".join(recent_events[-3:]) if recent_events else "Nothing notable recently"
# Try LLM generation first
if not self.llm.is_mock_mode:
try:
plot = await self._generate_llm_plot(
day=day,
weather=weather,
alive_count=alive_count,
agents_summary=agents_summary,
events_summary=events_summary,
tension_level=tension_level,
mood_avg=mood_avg,
)
if plot:
self._current_plot = plot
return plot
except Exception as e:
logger.error(f"LLM plot generation failed: {e}")
# Fallback to template-based generation
plot = self._generate_fallback_plot(weather, tension_level, mood_avg)
self._current_plot = plot
return plot
async def _generate_llm_plot(
self,
day: int,
weather: str,
alive_count: int,
agents_summary: str,
events_summary: str,
tension_level: str,
mood_avg: int,
) -> PlotPoint | None:
"""Generate plot point using LLM."""
# Build the prompt for the AI Director
system_prompt = f"""You are the AI Director for a survival drama on a deserted island.
Your role is to create dramatic narrative moments that engage the audience.
CURRENT SITUATION:
- Day {day} on the island
- Weather: {weather}
- Survivors: {alive_count} ({agents_summary})
- Recent events: {events_summary}
- Tension level: {tension_level}
- Average mood: {mood_avg}/100
RECENTLY USED PLOTS (avoid these):
{', '.join(self._plot_history) if self._plot_history else 'None yet'}
GUIDELINES:
1. Create dramatic tension appropriate to the tension level
2. Choices should have meaningful trade-offs
3. Consider weather and mood in your narrative
4. Keep descriptions cinematic but brief (under 50 words)
OUTPUT FORMAT (strict JSON):
{{
"title": "Brief dramatic title (3-5 words)",
"description": "Cinematic description of the situation (under 50 words)",
"choices": [
{{"id": "choice_a", "text": "First option (under 15 words)", "effects": {{"mood_delta": 5}}}},
{{"id": "choice_b", "text": "Second option (under 15 words)", "effects": {{"mood_delta": -5}}}}
]
}}"""
user_prompt = f"""The current tension is {tension_level}.
{"Create an intense, high-stakes event!" if tension_level == "high" else "Create an interesting event to raise the drama." if tension_level == "low" else "Create a moderately dramatic event."}
Generate a plot point now. Output ONLY valid JSON, no explanation."""
try:
# Use LLM service's internal acompletion
kwargs = {
"model": self.llm._model,
"messages": [
{"role": "system", "content": system_prompt},
{"role": "user", "content": user_prompt}
],
"max_tokens": 300,
"temperature": 0.9,
}
if self.llm._api_base:
kwargs["api_base"] = self.llm._api_base
if self.llm._api_key and not self.llm._api_key_header:
kwargs["api_key"] = self.llm._api_key
if self.llm._extra_headers:
kwargs["extra_headers"] = self.llm._extra_headers
response = await self.llm._acompletion(**kwargs)
content = response.choices[0].message.content.strip()
# Parse JSON response
# Handle potential markdown code blocks
if content.startswith("```"):
content = content.split("```")[1]
if content.startswith("json"):
content = content[4:]
data = json.loads(content)
# Validate and construct PlotPoint
choices = [
PlotChoice(
choice_id=c.get("id", f"choice_{i}"),
text=c.get("text", "Unknown option"),
effects=c.get("effects", {}),
)
for i, c in enumerate(data.get("choices", []))
]
if len(choices) < 2:
logger.warning("LLM returned fewer than 2 choices, using fallback")
return None
return PlotPoint(
plot_id=uuid.uuid4().hex,
title=data.get("title", "Unexpected Event"),
description=data.get("description", "Something happens..."),
choices=choices[:2], # Limit to 2 choices
ttl_seconds=60,
)
except json.JSONDecodeError as e:
logger.error(f"Failed to parse LLM JSON response: {e}")
return None
except Exception as e:
logger.error(f"LLM plot generation error: {e}")
return None
def _generate_fallback_plot(
self,
weather: str,
tension_level: str,
mood_avg: int,
) -> PlotPoint:
"""Generate plot point from templates when LLM is unavailable."""
# Filter templates based on context
available = [t for t in FALLBACK_PLOT_TEMPLATES if t["title"] not in self._plot_history]
if not available:
available = FALLBACK_PLOT_TEMPLATES
# Weight selection based on weather and tension
if weather.lower() in ("stormy", "rainy", "thunder"):
# Prefer storm-related plots
storm_plots = [t for t in available if "storm" in t["title"].lower()]
if storm_plots:
available = storm_plots
elif tension_level == "low" and mood_avg > 60:
# Prefer dramatic plots to shake things up
drama_plots = [t for t in available if "crisis" in t["title"].lower() or "trust" in t["title"].lower()]
if drama_plots:
available = drama_plots
template = self._rng.choice(available)
return PlotPoint(
plot_id=uuid.uuid4().hex,
title=template["title"],
description=template["description"],
choices=list(template["choices"]),
ttl_seconds=60,
)
async def resolve_vote(
self,
plot_point: PlotPoint,
winning_choice_id: str,
world_state: dict[str, Any],
) -> ResolutionResult:
"""
Resolve the vote and generate consequences.
Args:
plot_point: The PlotPoint that was voted on
winning_choice_id: The ID of the winning choice
world_state: Current world state for context
Returns:
ResolutionResult with message and effects to apply
"""
# Find the winning choice
winning_choice = next(
(c for c in plot_point.choices if c.choice_id == winning_choice_id),
plot_point.choices[0] # Fallback to first choice
)
# Try LLM resolution first
if not self.llm.is_mock_mode:
try:
result = await self._generate_llm_resolution(
plot_point=plot_point,
winning_choice=winning_choice,
world_state=world_state,
)
if result:
return result
except Exception as e:
logger.error(f"LLM resolution failed: {e}")
# Fallback resolution
return self._generate_fallback_resolution(plot_point, winning_choice)
async def _generate_llm_resolution(
self,
plot_point: PlotPoint,
winning_choice: PlotChoice,
world_state: dict[str, Any],
) -> ResolutionResult | None:
"""Generate resolution using LLM."""
agents_summary = ", ".join([
a.get("name", "Unknown")
for a in world_state.get("alive_agents", [])[:5]
]) or "the survivors"
system_prompt = f"""You are the AI Director narrating the consequences of an audience vote.
THE SITUATION:
{plot_point.description}
THE AUDIENCE VOTED FOR:
"{winning_choice.text}"
SURVIVORS INVOLVED:
{agents_summary}
GUIDELINES:
1. Describe the immediate consequence dramatically
2. Mention how the survivors react
3. Keep it brief but impactful (under 40 words)
4. The effects should feel meaningful
OUTPUT FORMAT (strict JSON):
{{
"message": "Dramatic description of what happens...",
"effects": {{
"mood_delta": -5,
"hp_delta": 0,
"energy_delta": -10,
"item_gained": null,
"item_lost": null,
"relationship_change": null
}}
}}"""
try:
kwargs = {
"model": self.llm._model,
"messages": [
{"role": "system", "content": system_prompt},
{"role": "user", "content": f"Narrate the consequence of choosing: {winning_choice.text}"}
],
"max_tokens": 200,
"temperature": 0.8,
}
if self.llm._api_base:
kwargs["api_base"] = self.llm._api_base
if self.llm._api_key and not self.llm._api_key_header:
kwargs["api_key"] = self.llm._api_key
if self.llm._extra_headers:
kwargs["extra_headers"] = self.llm._extra_headers
response = await self.llm._acompletion(**kwargs)
content = response.choices[0].message.content.strip()
# Handle markdown code blocks
if content.startswith("```"):
content = content.split("```")[1]
if content.startswith("json"):
content = content[4:]
data = json.loads(content)
# Merge LLM effects with choice's predefined effects
effects = {**winning_choice.effects, **data.get("effects", {})}
return ResolutionResult(
plot_id=plot_point.plot_id,
choice_id=winning_choice.choice_id,
message=data.get("message", f"The survivors chose: {winning_choice.text}"),
effects=effects,
)
except Exception as e:
logger.error(f"LLM resolution error: {e}")
return None
def _generate_fallback_resolution(
self,
plot_point: PlotPoint,
winning_choice: PlotChoice,
) -> ResolutionResult:
"""Generate fallback resolution message."""
# Template-based resolution messages
messages = [
f"The decision is made! {winning_choice.text}",
f"The survivors act: {winning_choice.text}",
f"Following the audience's choice: {winning_choice.text}",
]
return ResolutionResult(
plot_id=plot_point.plot_id,
choice_id=winning_choice.choice_id,
message=self._rng.choice(messages),
effects=dict(winning_choice.effects),
)
def calculate_tension_level(self, world_state: dict[str, Any]) -> str:
"""
Calculate the current tension level based on world state.
Args:
world_state: Dictionary with game state information
Returns:
"low", "medium", or "high"
"""
score = 0
# Factor: Agent health
alive_agents = world_state.get("alive_agents", [])
if alive_agents:
avg_hp = sum(a.get("hp", 100) for a in alive_agents) / len(alive_agents)
if avg_hp < 30:
score += 3
elif avg_hp < 50:
score += 2
elif avg_hp < 70:
score += 1
# Factor: Weather severity
weather = world_state.get("weather", "").lower()
if weather in ("stormy", "thunder"):
score += 2
elif weather in ("rainy",):
score += 1
# Factor: Mood
mood_avg = world_state.get("mood_avg", 50)
if mood_avg < 30:
score += 2
elif mood_avg < 50:
score += 1
# Factor: Recent deaths
recent_deaths = world_state.get("recent_deaths", 0)
score += min(recent_deaths * 2, 4)
# Factor: Low resources
if world_state.get("resources_critical", False):
score += 2
# Determine level
if score >= 6:
return "high"
elif score >= 3:
return "medium"
return "low"
# Global instance
director_service = DirectorService()

View File

@@ -15,6 +15,8 @@ from .database import init_db, get_db_session
from .models import User, Agent, WorldState, GameConfig, AgentRelationship
from .llm import llm_service
from .memory_service import memory_service
from .director_service import DirectorService, GameMode, PlotPoint
from .vote_manager import VoteManager, VoteOption, VoteSnapshot
if TYPE_CHECKING:
from .server import ConnectionManager
@@ -57,6 +59,14 @@ REVIVE_COST = 10 # Casual mode cost
INITIAL_USER_GOLD = 100
IDLE_CHAT_PROBABILITY = 0.15
# =============================================================================
# AI Director & Narrative Voting (Phase 9)
# =============================================================================
DIRECTOR_TRIGGER_INTERVAL = 60 # Ticks between narrative events (5 minutes at 5s/tick)
DIRECTOR_MIN_ALIVE_AGENTS = 2 # Minimum alive agents to trigger narrative
VOTING_DURATION_SECONDS = 60 # Duration of voting window
VOTE_BROADCAST_INTERVAL = 1.0 # How often to broadcast vote updates
# =============================================================================
# Day/Night cycle
# =============================================================================
@@ -136,6 +146,20 @@ class GameEngine:
# Key: agent_id (who needs to respond), Value: {partner_id, last_text, topic, expires_at_tick}
self._active_conversations = {}
# Phase 9: AI Director & Narrative Voting
self._director = DirectorService()
self._vote_manager = VoteManager(
duration_seconds=VOTING_DURATION_SECONDS,
broadcast_interval=VOTE_BROADCAST_INTERVAL,
)
self._game_mode = GameMode.SIMULATION
self._last_narrative_tick = 0
self._current_plot: PlotPoint | None = None
self._mode_change_tick = 0 # Tick when mode changed
# Set up vote broadcast callback
self._vote_manager.set_broadcast_callback(self._on_vote_update)
@property
def is_running(self) -> bool:
return self._running
@@ -250,6 +274,222 @@ class GameEngine:
if world:
await self._broadcast_event(EventType.WORLD_UPDATE, world.to_dict())
# =========================================================================
# AI Director & Narrative Voting (Phase 9)
# =========================================================================
async def _on_vote_update(self, snapshot: VoteSnapshot) -> None:
"""Callback for broadcasting vote updates."""
await self._broadcast_event(EventType.VOTE_UPDATE, snapshot.to_dict())
async def _set_game_mode(self, new_mode: GameMode, message: str = "") -> None:
"""Switch game mode and broadcast the change."""
old_mode = self._game_mode
self._game_mode = new_mode
self._mode_change_tick = self._tick_count
ends_at = 0.0
if new_mode == GameMode.VOTING:
session = self._vote_manager.current_session
if session:
ends_at = session.end_ts
await self._broadcast_event(EventType.MODE_CHANGE, {
"mode": new_mode.value,
"old_mode": old_mode.value,
"message": message,
"ends_at": ends_at,
})
logger.info(f"Game mode changed: {old_mode.value} -> {new_mode.value}")
def _get_world_state_for_director(self) -> dict:
"""Build world state context for the Director."""
with get_db_session() as db:
world = db.query(WorldState).first()
agents = db.query(Agent).filter(Agent.status == "Alive").all()
alive_agents = [
{"name": a.name, "hp": a.hp, "energy": a.energy, "mood": a.mood}
for a in agents
]
mood_avg = sum(a.mood for a in agents) / len(agents) if agents else 50
return {
"day": world.day_count if world else 1,
"weather": world.weather if world else "Sunny",
"time_of_day": world.time_of_day if world else "day",
"alive_agents": alive_agents,
"mood_avg": mood_avg,
"recent_events": [], # Could be populated from event history
"tension_level": self._director.calculate_tension_level({
"alive_agents": alive_agents,
"weather": world.weather if world else "Sunny",
"mood_avg": mood_avg,
}),
}
async def _should_trigger_narrative(self) -> bool:
"""Check if conditions are met to trigger a narrative event."""
# Only trigger in simulation mode
if self._game_mode != GameMode.SIMULATION:
return False
# Check tick interval
ticks_since_last = self._tick_count - self._last_narrative_tick
if ticks_since_last < DIRECTOR_TRIGGER_INTERVAL:
return False
# Check minimum alive agents
with get_db_session() as db:
alive_count = db.query(Agent).filter(Agent.status == "Alive").count()
if alive_count < DIRECTOR_MIN_ALIVE_AGENTS:
return False
return True
async def _trigger_narrative_event(self) -> None:
"""Trigger a narrative event from the Director."""
logger.info("Director triggering narrative event...")
# Switch to narrative mode
await self._set_game_mode(GameMode.NARRATIVE, "The Director intervenes...")
# Generate plot point
world_state = self._get_world_state_for_director()
plot = await self._director.generate_plot_point(world_state)
self._current_plot = plot
self._last_narrative_tick = self._tick_count
# Broadcast narrative event
await self._broadcast_event(EventType.NARRATIVE_PLOT, plot.to_dict())
logger.info(f"Narrative event: {plot.title}")
# Start voting session
options = [
VoteOption(choice_id=c.choice_id, text=c.text)
for c in plot.choices
]
self._vote_manager.start_vote(options, duration_seconds=VOTING_DURATION_SECONDS)
# Broadcast vote started
vote_data = self._vote_manager.get_vote_started_data()
if vote_data:
await self._broadcast_event(EventType.VOTE_STARTED, vote_data)
# Switch to voting mode
await self._set_game_mode(
GameMode.VOTING,
f"Vote now! {plot.choices[0].text} or {plot.choices[1].text}"
)
async def _process_voting_tick(self) -> None:
"""Process voting phase - check if voting has ended."""
if self._game_mode != GameMode.VOTING:
return
result = self._vote_manager.maybe_finalize()
if result:
# Voting ended
await self._broadcast_event(EventType.VOTE_ENDED, {
"vote_id": result.vote_id,
"total_votes": result.total_votes,
})
await self._broadcast_event(EventType.VOTE_RESULT, result.to_dict())
# Switch to resolution mode
await self._set_game_mode(
GameMode.RESOLUTION,
f"The audience has spoken: {result.winning_choice_text}"
)
# Process resolution
await self._process_vote_result(result)
async def _process_vote_result(self, result) -> None:
"""Process the voting result and apply consequences."""
if not self._current_plot:
logger.error("No current plot for resolution")
await self._set_game_mode(GameMode.SIMULATION, "Returning to normal...")
return
# Get resolution from Director
world_state = self._get_world_state_for_director()
resolution = await self._director.resolve_vote(
plot_point=self._current_plot,
winning_choice_id=result.winning_choice_id,
world_state=world_state,
)
# Apply effects
await self._apply_resolution_effects(resolution.effects)
# Broadcast resolution
await self._broadcast_event(EventType.RESOLUTION_APPLIED, resolution.to_dict())
logger.info(f"Resolution applied: {resolution.message}")
# Clear current plot
self._director.clear_current_plot()
self._current_plot = None
# Return to simulation after a brief pause
await asyncio.sleep(3.0) # Let players read the resolution
await self._set_game_mode(GameMode.SIMULATION, "The story continues...")
async def _apply_resolution_effects(self, effects: dict) -> None:
"""Apply resolution effects to the game world."""
def _coerce_int(value) -> int:
"""Safely convert LLM output (string/float/int) to int."""
try:
return int(value)
except (TypeError, ValueError):
return 0
mood_delta = _coerce_int(effects.get("mood_delta", 0))
hp_delta = _coerce_int(effects.get("hp_delta", 0))
energy_delta = _coerce_int(effects.get("energy_delta", 0))
if not any([mood_delta, hp_delta, energy_delta]):
return
with get_db_session() as db:
agents = db.query(Agent).filter(Agent.status == "Alive").all()
for agent in agents:
if mood_delta:
agent.mood = max(0, min(100, agent.mood + mood_delta))
if hp_delta:
agent.hp = max(0, min(100, agent.hp + hp_delta))
if energy_delta:
agent.energy = max(0, min(100, agent.energy + energy_delta))
logger.info(
f"Applied resolution effects: mood={mood_delta}, "
f"hp={hp_delta}, energy={energy_delta}"
)
async def process_vote(self, voter_id: str, choice_index: int, source: str = "twitch") -> bool:
"""
Process a vote from Twitch or Unity.
Args:
voter_id: Unique identifier for the voter
choice_index: 0-indexed choice number
source: Vote source ("twitch" or "unity")
Returns:
True if vote was recorded
"""
if self._game_mode != GameMode.VOTING:
return False
return self._vote_manager.cast_vote(voter_id, choice_index, source)
def parse_vote_command(self, message: str) -> int | None:
"""Parse a message for vote commands. Returns choice index or None."""
return self._vote_manager.parse_twitch_message(message)
# =========================================================================
# Day/Night cycle (Phase 2)
# =========================================================================
@@ -1702,6 +1942,20 @@ class GameEngine:
while self._running:
self._tick_count += 1
# Phase 9: Check voting phase (always runs)
await self._process_voting_tick()
# Phase 9: Check if we should trigger a narrative event
if await self._should_trigger_narrative():
await self._trigger_narrative_event()
# Skip simulation processing during narrative/voting/resolution modes
if self._game_mode != GameMode.SIMULATION:
await asyncio.sleep(self._tick_interval)
continue
# ========== SIMULATION MODE PROCESSING ==========
# 1. Advance time (Phase 2)
phase_change = await self._advance_time()
if phase_change:
@@ -1774,7 +2028,8 @@ class GameEngine:
"day": day,
"time_of_day": time_of_day,
"weather": weather,
"alive_agents": alive_count
"alive_agents": alive_count,
"game_mode": self._game_mode.value # Phase 9: Include game mode
})
await asyncio.sleep(self._tick_interval)

View File

@@ -66,6 +66,15 @@ class EventType(str, Enum):
VFX_EVENT = "vfx_event" # Visual effect trigger
GIFT_EFFECT = "gift_effect" # Twitch bits/sub effect
# AI Director & Narrative Voting (Phase 9)
MODE_CHANGE = "mode_change" # Game mode transition
NARRATIVE_PLOT = "narrative_plot" # Director generated plot point
VOTE_STARTED = "vote_started" # Voting session started
VOTE_UPDATE = "vote_update" # Real-time vote count update
VOTE_ENDED = "vote_ended" # Voting closed
VOTE_RESULT = "vote_result" # Final voting result
RESOLUTION_APPLIED = "resolution_applied" # Plot resolution executed
class GameEvent(BaseModel):
"""

View File

@@ -74,6 +74,18 @@ class TwitchBot(commands.Bot):
# Log the message for debugging
logger.info(f"Twitch chat [{username}]: {content}")
# Phase 9: Check for vote commands first (!1, !2, !A, !B)
vote_index = self._game_engine.parse_vote_command(content)
if vote_index is not None:
try:
voted = await self._game_engine.process_vote(username, vote_index, "twitch")
if voted:
logger.info(f"Vote recorded: {username} -> option {vote_index + 1}")
return # Don't process as regular command
except Exception as e:
logger.error(f"Error processing vote: {e}")
return
# Forward to game engine for command processing
try:
await self._game_engine.process_command(username, content)

445
backend/app/vote_manager.py Normal file
View File

@@ -0,0 +1,445 @@
"""
Vote Manager - Audience Voting System (Phase 9).
Manages voting sessions for narrative decisions,
supporting both Twitch chat commands and Unity client votes.
"""
from __future__ import annotations
import asyncio
import logging
import re
import time
import uuid
from dataclasses import dataclass, field
from typing import Any, Callable, Awaitable
logger = logging.getLogger(__name__)
@dataclass(frozen=True)
class VoteOption:
"""A voting option."""
choice_id: str
text: str
@dataclass
class VoteSession:
"""An active voting session."""
vote_id: str
options: list[VoteOption]
start_ts: float
end_ts: float
duration_seconds: int = 60 # Store actual duration for this session
votes_by_user: dict[str, int] = field(default_factory=dict) # user_id -> choice_index
tallies: list[int] = field(default_factory=list) # vote count per option
@dataclass(frozen=True)
class VoteSnapshot:
"""Real-time voting statistics snapshot."""
vote_id: str
tallies: list[int]
percentages: list[float]
total_votes: int
remaining_seconds: float
ends_at: float
def to_dict(self) -> dict[str, Any]:
"""Convert to dictionary for broadcasting."""
return {
"vote_id": self.vote_id,
"tallies": self.tallies,
"percentages": self.percentages,
"total_votes": self.total_votes,
"remaining_seconds": max(0, self.remaining_seconds),
"ends_at": self.ends_at,
}
@dataclass(frozen=True)
class VoteResult:
"""Final voting result after session ends."""
vote_id: str
winning_choice_id: str
winning_choice_text: str
winning_index: int
tallies: list[int]
percentages: list[float]
total_votes: int
is_tie: bool = False
def to_dict(self) -> dict[str, Any]:
"""Convert to dictionary for broadcasting."""
return {
"vote_id": self.vote_id,
"winning_choice_id": self.winning_choice_id,
"winning_choice_text": self.winning_choice_text,
"winning_index": self.winning_index,
"tallies": self.tallies,
"percentages": self.percentages,
"total_votes": self.total_votes,
"is_tie": self.is_tie,
}
# Twitch command patterns
VOTE_PATTERN_NUMERIC = re.compile(r"^!([1-9])$") # !1, !2, etc.
VOTE_PATTERN_ALPHA = re.compile(r"^!([AaBb])$") # !A, !B, etc.
class VoteManager:
"""
Manages voting sessions with dual-channel support (Twitch + Unity).
Features:
- Real-time vote counting
- Vote changing (users can change their vote)
- Automatic session expiration
- Periodic snapshot broadcasting
"""
def __init__(
self,
duration_seconds: int = 60,
broadcast_interval: float = 1.0,
) -> None:
"""
Initialize the vote manager.
Args:
duration_seconds: Default voting window duration
broadcast_interval: How often to broadcast vote updates (seconds)
"""
self._duration_seconds = duration_seconds
self._broadcast_interval = broadcast_interval
self._current: VoteSession | None = None
self._broadcast_callback: Callable[[VoteSnapshot], Awaitable[None]] | None = None
self._broadcast_task: asyncio.Task | None = None
@property
def is_voting_active(self) -> bool:
"""Check if a voting session is currently active."""
if not self._current:
return False
return time.time() < self._current.end_ts
@property
def current_session(self) -> VoteSession | None:
"""Get the current voting session."""
return self._current
def set_broadcast_callback(
self,
callback: Callable[[VoteSnapshot], Awaitable[None]],
) -> None:
"""
Set the callback for broadcasting vote updates.
Args:
callback: Async function that receives VoteSnapshot and broadcasts it
"""
self._broadcast_callback = callback
def start_vote(
self,
options: list[VoteOption],
duration_seconds: int | None = None,
now: float | None = None,
) -> VoteSession:
"""
Start a new voting session.
Args:
options: List of voting options (minimum 2)
duration_seconds: Override default duration
now: Override current timestamp (for testing)
Returns:
The created VoteSession
"""
if len(options) < 2:
raise ValueError("Voting requires at least 2 options")
now = now or time.time()
duration = duration_seconds or self._duration_seconds
session = VoteSession(
vote_id=uuid.uuid4().hex,
options=options,
start_ts=now,
end_ts=now + duration,
duration_seconds=duration,
tallies=[0 for _ in options],
)
self._current = session
# Start broadcast loop
if self._broadcast_callback:
self._start_broadcast_loop()
logger.info(
f"Vote started: {session.vote_id} with {len(options)} options, "
f"duration={duration}s"
)
return session
def _start_broadcast_loop(self) -> None:
"""Start the periodic broadcast task."""
if self._broadcast_task and not self._broadcast_task.done():
self._broadcast_task.cancel()
async def broadcast_loop():
try:
while self.is_voting_active:
snapshot = self.snapshot()
if snapshot and self._broadcast_callback:
try:
await self._broadcast_callback(snapshot)
except Exception as e:
logger.error(f"Broadcast callback error: {e}")
await asyncio.sleep(self._broadcast_interval)
except asyncio.CancelledError:
pass
self._broadcast_task = asyncio.create_task(broadcast_loop())
def parse_twitch_message(self, content: str) -> int | None:
"""
Parse a Twitch chat message for vote commands.
Supported formats:
- !1, !2, !3, etc. (1-indexed, converted to 0-indexed)
- !A, !B (converted to 0, 1)
Args:
content: The chat message content
Returns:
Choice index (0-indexed) or None if not a vote command
"""
text = content.strip()
# Try numeric pattern first
match = VOTE_PATTERN_NUMERIC.match(text)
if match:
return int(match.group(1)) - 1 # Convert to 0-indexed
# Try alphabetic pattern
match = VOTE_PATTERN_ALPHA.match(text)
if match:
letter = match.group(1).upper()
return ord(letter) - ord('A') # A=0, B=1
return None
def cast_vote(
self,
voter_id: str,
choice_index: int,
source: str = "twitch",
) -> bool:
"""
Record a vote from a user.
Users can change their vote - the previous vote is subtracted
and the new vote is added.
Args:
voter_id: Unique identifier for the voter
choice_index: 0-indexed choice number
source: Vote source ("twitch" or "unity")
Returns:
True if vote was recorded, False if invalid or session ended
"""
if not self._current:
logger.debug(f"Vote rejected: no active session (voter={voter_id})")
return False
if time.time() > self._current.end_ts:
logger.debug(f"Vote rejected: session ended (voter={voter_id})")
return False
if choice_index < 0 or choice_index >= len(self._current.options):
logger.debug(
f"Vote rejected: invalid choice {choice_index} "
f"(voter={voter_id}, max={len(self._current.options)-1})"
)
return False
# Normalize voter ID (Twitch usernames are case-insensitive)
normalized_voter_id = voter_id.strip().lower()
if not normalized_voter_id:
logger.debug("Vote rejected: empty voter id")
return False
# Handle vote change - subtract previous vote
previous = self._current.votes_by_user.get(normalized_voter_id)
if previous is not None:
if previous == choice_index:
# Same vote, no change needed
return True
# Subtract old vote
self._current.tallies[previous] = max(
0, self._current.tallies[previous] - 1
)
logger.debug(f"Vote changed: {normalized_voter_id} from {previous} to {choice_index}")
# Record new vote
self._current.votes_by_user[normalized_voter_id] = choice_index
self._current.tallies[choice_index] += 1
logger.debug(
f"Vote cast: {voter_id} -> {choice_index} "
f"(source={source}, tallies={self._current.tallies})"
)
return True
def snapshot(self, now: float | None = None) -> VoteSnapshot | None:
"""
Generate a real-time snapshot of current voting status.
Args:
now: Override current timestamp (for testing)
Returns:
VoteSnapshot or None if no active session
"""
if not self._current:
return None
now = now or time.time()
tallies = list(self._current.tallies)
total = sum(tallies)
# Calculate percentages
if total > 0:
percentages = [round((t / total) * 100, 1) for t in tallies]
else:
percentages = [0.0 for _ in tallies]
return VoteSnapshot(
vote_id=self._current.vote_id,
tallies=tallies,
percentages=percentages,
total_votes=total,
remaining_seconds=self._current.end_ts - now,
ends_at=self._current.end_ts,
)
def maybe_finalize(self, now: float | None = None) -> VoteResult | None:
"""
Check if voting has ended and finalize results.
Args:
now: Override current timestamp (for testing)
Returns:
VoteResult if voting ended, None if still active
"""
if not self._current:
return None
now = now or time.time()
if now < self._current.end_ts:
return None
# Cancel broadcast task
if self._broadcast_task and not self._broadcast_task.done():
self._broadcast_task.cancel()
# Calculate final results
tallies = list(self._current.tallies)
total = sum(tallies)
# Calculate percentages
if total > 0:
percentages = [round((t / total) * 100, 1) for t in tallies]
else:
percentages = [0.0 for _ in tallies]
# Find winner
if tallies:
max_votes = max(tallies)
winners = [i for i, t in enumerate(tallies) if t == max_votes]
is_tie = len(winners) > 1
# In case of tie, choose randomly (or could defer to Director)
import random
winning_index = random.choice(winners) if is_tie else winners[0]
else:
winning_index = 0
is_tie = False
winning_option = self._current.options[winning_index]
result = VoteResult(
vote_id=self._current.vote_id,
winning_choice_id=winning_option.choice_id,
winning_choice_text=winning_option.text,
winning_index=winning_index,
tallies=tallies,
percentages=percentages,
total_votes=total,
is_tie=is_tie,
)
logger.info(
f"Vote finalized: {result.vote_id} "
f"winner={result.winning_choice_id} ({result.winning_choice_text}) "
f"votes={result.tallies} tie={result.is_tie}"
)
# Clear current session
self._current = None
return result
def cancel_vote(self) -> bool:
"""
Cancel the current voting session.
Returns:
True if a session was cancelled, False if no active session
"""
if not self._current:
return False
if self._broadcast_task and not self._broadcast_task.done():
self._broadcast_task.cancel()
vote_id = self._current.vote_id
self._current = None
logger.info(f"Vote cancelled: {vote_id}")
return True
def get_vote_started_data(self) -> dict[str, Any] | None:
"""
Get data for VOTE_STARTED event.
Returns:
Dictionary with vote session info, or None if no active session
"""
if not self._current:
return None
return {
"vote_id": self._current.vote_id,
"choices": [
{"choice_id": o.choice_id, "text": o.text}
for o in self._current.options
],
"duration_seconds": self._current.duration_seconds,
"ends_at": self._current.end_ts,
"source": "director",
}
# Global instance
vote_manager = VoteManager()