Traitorsim Orchestrator

TraitorSim Orchestrator

Coordinate complete TraitorSim workflows by combining multiple specialized skills. This orchestrator guides you through persona generation, game configuration, execution, and post-game analysis.

Quick Start

Complete end-to-end workflow:

# 1. Generate persona library (one-time setup)
/persona-pipeline --count 50

# 2. Configure simulation
/simulation-config --rule-set UK --players 22

# 3. Run game
python -m src.traitorsim

# 4. Analyze results
/game-analyzer --game-log data/logs/latest.json

Available Workflows

Workflow 1: Persona Library Creation (One-Time Setup)

Goal: Generate a reusable library of 50-100 personas

Skills used:

1. archetype-designer - Design or review archetypes
2. quota-manager - Plan API quota usage
3. persona-pipeline - Generate personas via Deep Research + Claude
4. world-bible-validator - Validate lore consistency

Steps:

# Step 1: Review archetype definitions
/archetype-designer
# Inspect 13 archetypes, adjust OCEAN ranges if needed

# Step 2: Plan quota usage
/quota-manager
# For 50 personas: ~$20-25, ~6-8 hours with quota limits
# Decide on wave strategy (6→4→2→2 pattern)

# Step 3: Generate persona library
/persona-pipeline --count 50
# Runs 5-stage pipeline:
#   - Generate skeletons
#   - Submit Deep Research jobs (in waves)
#   - Poll until complete
#   - Synthesize backstories with Claude Opus
#   - Validate all personas

# Step 4: Validate World Bible compliance
/world-bible-validator --library data/personas/library/production_50_personas.json
# Check for forbidden brand leakage
# Verify in-universe brand usage

# Output: data/personas/library/production_50_personas.json

Timeline:

• Archetype review: 30 min
• Skeleton generation: 5 min
• Deep Research submission + polling: 2-4 hours (quota-limited)
• Synthesis: 15-30 min
• Validation: 5 min
• Total: ~3-5 hours active work, 6-8 hours elapsed

Cost: $20-25 for 50 personas ($0.40-0.50 each)

Workflow 2: Run Single Simulation

Goal: Run one game with existing persona library

Skills used:

1. simulation-config - Set game rules and parameters
2. (Run game via main script)
3. memory-debugger - Inspect agent behaviors if issues
4. game-analyzer - Analyze outcomes and patterns

Steps:

# Step 1: Configure simulation
/simulation-config
# Choose: UK/US/Australia rules
# Set player count, Traitor count, recruitment type

# Example: Standard UK game
python -c "
from src.traitorsim.core.config import SimulationConfig
config = SimulationConfig(
    rule_set='UK',
    num_players=22,
    num_traitors=4,
    persona_library_path='data/personas/library/production_50_personas.json'
)
config.save('configs/uk_standard.json')
"

# Step 2: Run simulation
python -m src.traitorsim --config configs/uk_standard.json

# Step 3: If issues arise, debug agent memory
/memory-debugger --player player_03
# Inspect profile.md, trust matrix, diary entries

# Step 4: Analyze game results
/game-analyzer --game-log data/logs/game_2025_12_21.json
# Trust matrix evolution
# Voting patterns
# Mission performance
# Emergent behaviors

# Output: Game log, analysis report

Timeline: 10-30 minutes per game (depends on agent count)

Cost: ~$2-5 per game (GameMaster + agent API calls)

Workflow 3: Batch Simulation for Research

Goal: Run 50+ games to analyze rule variants or archetype balance

Skills used:

1. simulation-config - Create multiple configurations
2. (Batch execution script)
3. game-analyzer - Aggregate analysis across games

Steps:

# Step 1: Create configurations
/simulation-config
# Generate 3 configs:
#   - UK standard
#   - UK with ultimatum recruitment
#   - UK with no recruitment

python -c "
from src.traitorsim.core.config import SimulationConfig

configs = [
    SimulationConfig(rule_set='UK', recruitment_type='standard'),
    SimulationConfig(rule_set='UK', recruitment_type='ultimatum'),
    SimulationConfig(rule_set='UK', recruitment_type='none')
]

for i, config in enumerate(configs):
    config.save(f'configs/experiment_{i}.json')
"

# Step 2: Run batch simulations
for i in {0..2}; do
    for trial in {1..50}; do
        python -m src.traitorsim --config configs/experiment_$i.json \
            --log-file data/logs/exp_${i}_trial_${trial}.json
    done
done

# Step 3: Aggregate analysis
/game-analyzer --batch
python scripts/aggregate_analysis.py \
    --input "data/logs/exp_*.json" \
    --output analysis/recruitment_experiment.md

# Analyze:
# - Traitor win rate by recruitment type
# - Average game length
# - Recruitment success rate

Timeline: 8-24 hours for 150 games (50 per config)

Cost: ~$300-750 for 150 games

Workflow 4: Incremental Persona Expansion

Goal: Add more personas to existing library without regenerating all

Skills used:

1. persona-pipeline (incremental mode)
2. quota-manager
3. world-bible-validator

Steps:

# Step 1: Check existing library
cat data/personas/library/production_50_personas.json | jq 'length'
# Output: 50

# Step 2: Plan quota for expansion
/quota-manager
# Adding 25 personas: ~$10-12, ~3-4 hours

# Step 3: Generate new personas (incremental)
/persona-pipeline --count 25 --incremental
# Pipeline automatically:
#   - Loads existing library
#   - Generates only NEW skeletons (avoiding duplicates)
#   - Synthesizes only NEW personas
#   - Merges with existing library

# Step 4: Validate merged library
/world-bible-validator --library data/personas/library/production_75_personas.json

# Output: data/personas/library/production_75_personas.json (75 total)

Timeline: ~2-4 hours Cost: ~$10-12 for 25 additional personas

Workflow 5: Debug Poor Game Outcome

Goal: Understand why a game had unexpected results

Skills used:

1. game-analyzer - Identify what happened
2. memory-debugger - Inspect agent states
3. simulation-config - Check if config was correct

Steps:

# Symptom: Traitors won too easily

# Step 1: Analyze game log
/game-analyzer --game-log data/logs/poor_game.json

# Check:
# - Were Traitors too powerful? (too many Traitors initially)
# - Did Faithfuls update trust matrices?
# - Were voting patterns logical?

# Step 2: Debug agent memory
/memory-debugger

# For each Faithful who performed poorly:
cat data/memories/player_05/suspects.csv
# Check if trust matrix updated at all

cat data/memories/player_05/diary/day_03_roundtable.md
# Check if observations were detailed

# Step 3: Review configuration
/simulation-config
cat configs/current_config.json

# Check:
# - Was num_traitors too high?
# - Was tie_break_method favoring Traitors?
# - Were archetypes balanced?

# Step 4: Identify root cause
# Examples:
# - Traitors = 30% of players (too high, should be 15-20%)
# - Trust matrices not updating (bug in memory manager)
# - All Faithfuls had low Openness (didn't update beliefs)

Workflow 6: New Archetype Development

Goal: Create and test a new archetype

Skills used:

1. archetype-designer - Define new archetype
2. persona-pipeline - Generate test personas
3. simulation-config - Run test games
4. game-analyzer - Validate archetype behavior

Steps:

# Step 1: Design new archetype
/archetype-designer

# Example: "The Paranoid Investigator"
python -c "
from src.traitorsim.core.archetypes import ArchetypeDefinition, ARCHETYPES

paranoid_investigator = ArchetypeDefinition(
    id='paranoid_investigator',
    name='The Paranoid Investigator',
    ocean_ranges={
        'openness': (0.65, 0.85),
        'conscientiousness': (0.70, 0.90),
        'extraversion': (0.35, 0.55),
        'agreeableness': (0.30, 0.50),
        'neuroticism': (0.75, 0.95)
    },
    # ... rest of archetype definition
)

ARCHETYPES['paranoid_investigator'] = paranoid_investigator
"

# Step 2: Generate test personas with new archetype
/persona-pipeline --archetype paranoid_investigator --count 3

# Step 3: Run test games with new archetype
/simulation-config
# Set up game with mix of archetypes including 2-3 paranoid investigators

python -m src.traitorsim --config configs/test_new_archetype.json

# Step 4: Analyze archetype behavior
/game-analyzer --focus-archetype paranoid_investigator

# Check:
# - Did high Neuroticism make them defensive?
# - Did high Conscientiousness improve trust tracking?
# - Did low Agreeableness lead to confrontations?
# - Was archetype balanced (not too powerful/weak)?

Skill Coordination Reference

When to Use Which Skill

Starting a new TraitorSim project: → Use traitorsim-orchestrator (this skill) → Follow Workflow 1

Creating character archetypes: → Use archetype-designer

• Define OCEAN trait ranges
• Set stat biases and demographics
• Design gameplay profiles

Generating personas: → Use persona-pipeline

• Run 5-stage pipeline (skeleton → research → synthesis → validation)
• Handle quota limits with wave submission
• Incremental generation for expansions

Managing API quotas: → Use quota-manager

• Client-side tracking
• Exponential backoff
• Wave-based submission strategies

Validating lore consistency: → Use world-bible-validator

• Detect forbidden brand leakage
• Ensure in-universe brand usage
• Batch validation reports

Configuring simulations: → Use simulation-config

• Set regional rules (UK/US/Australia)
• Configure player counts and recruitment mechanics
• Test rule variants

Debugging agent behavior: → Use memory-debugger

• Inspect profile.md, trust matrices, diary entries
• Validate memory updates
• Check skill files

Analyzing game outcomes: → Use game-analyzer

• Trust matrix evolution
• Voting pattern analysis
• Emergent behavior detection
• Personality-behavior correlation

Skill Dependencies

graph TD
    A[archetype-designer] --> B[persona-pipeline]
    C[quota-manager] -.-> B
    B --> D[world-bible-validator]
    D --> E[simulation-config]
    E --> F[Run Game]
    F --> G[memory-debugger]
    F --> H[game-analyzer]
    H -.-> G

    style A fill:#e1f5ff
    style B fill:#e1f5ff
    style C fill:#fff4e1
    style D fill:#e1ffe1
    style E fill:#ffe1f5
    style F fill:#f0f0f0
    style G fill:#ffe1e1
    style H fill:#ffe1e1

Legend:

• Blue: Persona creation pipeline
• Yellow: Support/optimization
• Green: Validation
• Pink: Game configuration
• Gray: Game execution
• Red: Debugging/analysis

Common Orchestrated Workflows

Full Production Setup (First Time)

# 1. Review/customize archetypes
/archetype-designer
# Review 13 default archetypes, create custom ones if needed

# 2. Generate production persona library
/persona-pipeline --count 100
# ~$40-50, 10-15 hours with quota limits

# 3. Validate library
/world-bible-validator --library data/personas/library/production_100_personas.json

# 4. Create default configs
/simulation-config
# Generate configs for UK, US, Australia variants

# 5. Run test games
for ruleset in UK US Australia; do
    python -m src.traitorsim --config configs/${ruleset}_standard.json
done

# 6. Analyze test games
/game-analyzer --batch data/logs/test_*.json

# 7. Adjust configs based on results
# If needed, regenerate specific archetypes or configs

# Production ready!

Daily Development Iteration

# 1. Make code changes to agent logic

# 2. Run quick test with small game
/simulation-config --players 10 --traitors 2
python -m src.traitorsim --config configs/dev_test.json

# 3. Debug if issues
/memory-debugger --player player_03
/game-analyzer --game-log data/logs/latest.json

# 4. Fix issues, repeat

Weekly Research Experiment

# Monday: Design experiment
/simulation-config
# Create 3-5 configs varying one parameter

# Tuesday-Thursday: Run batch simulations
# 50 games per config = 150-250 total games
# Automated batch script

# Friday: Analysis
/game-analyzer --batch
# Aggregate statistics
# Generate research report

# Present findings!

Troubleshooting Workflows

Problem: Personas have brand leakage

Workflow:

1. /world-bible-validator - Identify leaked brands
2. Update synthesis prompt in scripts/synthesize_backstories.py
3. /persona-pipeline --regenerate - Regenerate affected personas
4. /world-bible-validator - Re-check

Problem: Agents not updating trust matrices

Workflow:

1. /game-analyzer - Confirm trust matrices are static
2. /memory-debugger - Check if suspects.csv is being written
3. Review src/traitorsim/memory/memory_manager.py
4. Fix memory update logic
5. Re-run test game
6. /game-analyzer - Verify trust updates now occur

Problem: Traitors winning too often

Workflow:

1. /game-analyzer --batch - Calculate Traitor win rate across games
2. /simulation-config - Check Traitor % (should be 15-20%)
3. Adjust num_traitors or recruitment_type
4. Run 20 test games with new config
5. /game-analyzer --batch - Re-calculate win rate
6. Iterate until balanced (~40-60% Traitor win rate)

Problem: Quota errors during persona generation

Workflow:

1. /quota-manager - Review quota strategies
2. Update scripts/batch_deep_research.py with wave submission
3. Implement client-side QuotaTracker
4. /persona-pipeline --resume - Resume from last successful job
5. Monitor quota usage with logging

Best Practices

For Persona Generation Projects

1. Start small, scale up:

   • Generate 10-15 test personas first
   • Validate quality and cost
   • Then scale to 50-100 production personas

2. Use incremental generation:

   • Don't regenerate entire library for additions
   • Use --incremental flag to merge new personas

3. Monitor quotas closely:

   • Use QuotaTracker for large batches
   • Submit in waves (6→4→2→2 pattern)
   • Budget 1 hour per 3-5 personas for Deep Research

4. Validate early and often:

   • Run validation after each synthesis batch
   • Catch brand leakage immediately
   • Fix prompts before generating more

For Game Execution Projects

1. Test with small games first:

   • 10-12 players for quick iteration
   • Identify bugs faster

2. Use consistent configs:

   • Save configs to JSON files
   • Version control them
   • Reproduce issues with same config

3. Log everything:

   • Enable detailed game logs
   • Save agent memory states at each phase
   • Makes post-game debugging easier

4. Analyze failures immediately:

   • Don't batch 100 games before analyzing
   • Check first 5-10 games
   • Iterate on configs

When to Use This Skill

Use this skill when:

• Starting a new TraitorSim project from scratch
• Running complete end-to-end workflows
• Coordinating multiple specialized skills
• Troubleshooting complex issues spanning multiple components
• Planning multi-day research experiments
• Onboarding new team members to TraitorSim workflows

When NOT to Use This Skill

Don't use this skill for:

• Single-skill tasks (use the specific skill directly)
• Quick one-off queries about specific components
• Simple debugging of isolated issues

Instead, use the specialized skills directly for focused tasks.