infinite-agents-public/infinite_variants/infinite_variant_5/docs/configuration_guide.md

Configuration Guide

Complete reference for the Configuration-Driven Infinite Loop Orchestration system.

Table of Contents

1. Introduction
2. Configuration Architecture
3. Configuration Sections Reference
4. Profiles
5. Chain Prompting Configuration
6. Validation
7. Examples
8. Best Practices
9. Troubleshooting

Introduction

The configuration system externalizes all orchestration parameters to JSON files, enabling:

• Reproducibility: Save and share configurations for consistent results
• Flexibility: Adjust any parameter without code changes
• Quality: Multi-stage validation ensures correctness
• Scalability: Different profiles optimize for different scales
• Collaboration: Team members can share and compare configurations

Configuration Flow

User Command
    ↓
Load defaults.json
    ↓
Merge profile.json (if specified)
    ↓
Merge custom config (if provided)
    ↓
Apply inline overrides (if provided)
    ↓
Validate against schema
    ↓
Validate semantics
    ↓
Cross-field validation
    ↓
Execute with final config

Configuration Architecture

Hierarchical Merging

Configurations merge in this order (later overrides earlier):

1. Defaults (.claude/config/defaults.json)

   • Base configuration
   • Balanced settings for general use
   • All required fields present

2. Profile (.claude/config/profiles/{profile}.json)

   • Optional profile override
   • Optimized for specific use case
   • Only overrides necessary fields

3. Custom File (path provided in command)

   • User-specific configuration
   • Can override any field
   • Validated before merge

4. Inline Overrides (JSON string in command)

   • Runtime parameter changes
   • Highest priority
   • Useful for one-off adjustments

Deep Merge Strategy

The merge is deep (property-level), not shallow (object-level):

// Base
{
  "orchestration": {
    "max_parallel_agents": 3,
    "batch_size": 5
  }
}

// Override
{
  "orchestration": {
    "max_parallel_agents": 8
  }
}

// Result (max_parallel_agents overridden, batch_size preserved)
{
  "orchestration": {
    "max_parallel_agents": 8,
    "batch_size": 5
  }
}

Configuration Sections Reference

orchestration

Controls parallel execution and agent coordination.

{
  "orchestration": {
    "max_parallel_agents": 3,
    "batch_size": 5,
    "infinite_mode_wave_size": 5,
    "context_budget_per_agent": 50000,
    "agent_timeout_ms": 300000,
    "enable_progressive_sophistication": true
  }
}

max_parallel_agents

• Type: integer
• Range: 1-10
• Default: 3
• Description: Maximum number of sub-agents running simultaneously
• Impact:
  • Higher = faster execution but more context usage
  • Lower = slower but more careful generation
• Recommendations:
  • Development: 2-3
  • Production: 5-8
  • Research: 3-4

batch_size

• Type: integer
• Range: 1-50
• Default: 5
• Description: Number of iterations per generation batch
• Impact:
  • Higher = fewer waves, more efficient
  • Lower = more waves, better control
• Recommendations:
  • Testing: 3-5
  • Production: 10-20
  • Large batches: 20-50

infinite_mode_wave_size

• Type: integer
• Range: 1-20
• Default: 5
• Description: Iterations per wave in infinite mode
• Impact: Controls wave boundaries for progressive sophistication
• Recommendations:
  • Should be multiple of max_parallel_agents for efficiency
  • Typical: 5-10

context_budget_per_agent

• Type: integer
• Range: 10000-200000
• Default: 50000
• Description: Token budget allocated to each sub-agent
• Impact:
  • Higher = more room for context, web content, examples
  • Lower = risk of context exhaustion
• Recommendations:
  • Simple tasks: 30000-50000
  • Complex tasks: 60000-100000
  • Web-enhanced: 80000-120000

agent_timeout_ms

• Type: integer
• Range: 30000-600000 (30 seconds to 10 minutes)
• Default: 300000 (5 minutes)
• Description: Maximum execution time for each agent
• Impact: Prevents stuck agents from blocking execution
• Recommendations:
  • Simple generation: 120000-180000 (2-3 minutes)
  • Complex generation: 300000-420000 (5-7 minutes)
  • With review stage: 420000-600000 (7-10 minutes)

enable_progressive_sophistication

• Type: boolean
• Default: true
• Description: Increase complexity and sophistication in later waves
• Impact: Creates learning curve in infinite mode
• Use when: Running infinite mode for progressive improvement

generation

Controls output file creation and formatting.

{
  "generation": {
    "output_directory": "output",
    "naming_pattern": "{theme}_{iteration:03d}_{variant}.html",
    "file_format": "html",
    "include_metadata": true,
    "metadata_format": "html_comment"
  }
}

output_directory

• Type: string
• Default: "output"
• Description: Directory for generated files (relative to project root)
• Examples: "output", "output_dev", "visualizations", "generated/batch1"

naming_pattern

• Type: string
• Default: "{theme}{iteration:03d}{variant}.html"
• Description: File naming pattern with placeholders
• Placeholders:
  • {theme} - Content theme (from spec or agent)
  • {iteration} - Iteration number
  • {iteration:03d} - Zero-padded to 3 digits (001, 002, etc.)
  • {iteration:04d} - Zero-padded to 4 digits (0001, 0002, etc.)
  • {variant} - Variant descriptor
  • {timestamp} - ISO timestamp
• Examples:
  • viz_{iteration:04d}.html → viz_0001.html, viz_0002.html
  • {theme}_{timestamp}.html → climate_2025-10-10T14:30:00Z.html
  • output_{iteration}.html → output_1.html, output_2.html

file_format

• Type: string
• Enum: html, markdown, json, text
• Default: html
• Description: Output file format
• Impact: Determines file extension and content structure

include_metadata

• Type: boolean
• Default: true
• Description: Embed generation metadata in output files
• Impact: Adds comments with iteration info, web sources, techniques used

metadata_format

• Type: string
• Enum: html_comment, yaml_frontmatter, json_header
• Default: html_comment
• Description: Format for embedded metadata
• Examples:
  • html_comment: <!-- Metadata: ... -->
  • yaml_frontmatter: ---\nkey: value\n---
  • json_header: /* { "key": "value" } */

quality

Controls quality standards and validation.

{
  "quality": {
    "min_uniqueness_threshold": 0.85,
    "enable_validation": true,
    "enable_review_stage": false,
    "max_retry_attempts": 2,
    "require_spec_compliance_check": true
  }
}

min_uniqueness_threshold

• Type: number
• Range: 0.0-1.0
• Default: 0.85
• Description: Minimum required uniqueness score (0=identical, 1=completely unique)
• Impact: Controls iteration diversity
• Recommendations:
  • Permissive: 0.5-0.7 (allows similar variations)
  • Moderate: 0.7-0.85 (balanced uniqueness)
  • Strict: 0.85-0.95 (very unique)
  • Extreme: 0.95-1.0 (maximally unique)

enable_validation

• Type: boolean
• Default: true
• Description: Validate generated outputs against spec and config
• Impact: Catches errors, enforces quality, but adds execution time
• Use when: Quality is important, willing to trade time for correctness

enable_review_stage

• Type: boolean
• Default: false
• Description: Add chain prompting review stage after generation
• Impact:
  • Significantly improves quality (review + refine)
  • Increases execution time by 30-50%
  • Enables self-correction loops
• Use when: Maximum quality needed, time is not critical

max_retry_attempts

• Type: integer
• Range: 0-5
• Default: 2
• Description: Maximum retries for failed generations
• Impact: Improves success rate but increases total time
• Recommendations:
  • No retries: 0 (fail fast)
  • Normal: 2-3
  • Persistent: 4-5

require_spec_compliance_check

• Type: boolean
• Default: true
• Description: Validate outputs match specification requirements
• Impact: Ensures spec adherence, catches structural errors
• Use when: Spec has strict requirements

web_enhancement

Controls web-based learning and content fetching.

{
  "web_enhancement": {
    "enabled": false,
    "initial_priming_urls": 3,
    "urls_per_iteration": 1,
    "progressive_difficulty": true,
    "enable_web_search_fallback": true,
    "cache_web_content": false,
    "web_fetch_timeout_ms": 30000
  }
}

enabled

• Type: boolean
• Default: false
• Description: Enable web-enhanced generation with URL fetching
• Impact: Agents fetch web resources to inform generation
• Use when: Want to incorporate web knowledge, documentation, tutorials

initial_priming_urls

• Type: integer
• Range: 0-10
• Default: 3
• Description: Number of URLs to fetch during initial priming phase
• Impact: Builds foundational knowledge before generation starts
• Recommendations:
  • Quick start: 2-3
  • Moderate: 3-5
  • Comprehensive: 5-8
  • Maximum: 8-10

urls_per_iteration

• Type: integer
• Range: 0-5
• Default: 1
• Description: URLs to fetch per iteration during generation
• Impact: More URLs = more diverse learning but slower execution
• Recommendations:
  • Focused: 1
  • Balanced: 2
  • Diverse: 3-5

progressive_difficulty

• Type: boolean
• Default: true
• Description: Use progressive URL difficulty (foundation → expert)
• Impact: Early iterations use basic resources, later use advanced
• Use when: Want learning curve in generated content

enable_web_search_fallback

• Type: boolean
• Default: true
• Description: Use web search when pre-defined URLs exhausted
• Impact: Ensures continuous web learning in large batches
• Use when: Running many iterations, may exhaust URL list

cache_web_content

• Type: boolean
• Default: false
• Description: Cache fetched web content for reuse
• Impact:
  • Faster execution (no re-fetching)
  • Lower network usage
  • But may miss updated content
• Use when: Development, testing, same URLs used repeatedly

web_fetch_timeout_ms

• Type: integer
• Range: 5000-60000 (5-60 seconds)
• Default: 30000 (30 seconds)
• Description: Timeout for web fetch operations
• Impact: Prevents slow fetches from blocking execution
• Recommendations:
  • Fast resources: 10000-20000
  • Normal: 20000-30000
  • Slow/complex pages: 30000-45000

logging

Controls logging verbosity and output.

{
  "logging": {
    "level": "info",
    "log_agent_outputs": true,
    "log_web_fetches": true,
    "log_config_loading": true,
    "verbose": false
  }
}

level

• Type: string
• Enum: debug, info, warn, error
• Default: info
• Description: Minimum logging level
• Levels:
  • debug: All messages (very detailed)
  • info: Informational messages and above
  • warn: Warnings and errors only
  • error: Errors only
• Recommendations:
  • Development: debug
  • Testing: info
  • Production: warn or error

log_agent_outputs

• Type: boolean
• Default: true
• Description: Log what each sub-agent generates
• Impact: Useful for debugging but produces significant output
• Use when: Need to understand what agents are producing

log_web_fetches

• Type: boolean
• Default: true
• Description: Log web fetch operations and URLs
• Impact: Tracks web learning sources
• Use when: Need to audit web sources or debug fetch issues

log_config_loading

• Type: boolean
• Default: true
• Description: Log configuration loading and merging process
• Impact: Shows which configs loaded and how they merged
• Use when: Debugging configuration issues

verbose

• Type: boolean
• Default: false
• Description: Maximum verbosity (overrides level to debug)
• Impact: Extremely detailed output, use only for debugging
• Use when: Debugging complex issues, understanding system behavior

chain_prompting

Controls chain prompting workflow execution.

{
  "chain_prompting": {
    "enabled": true,
    "stages": [
      "load_config",
      "validate_config",
      "merge_config",
      "analyze_spec",
      "plan_execution",
      "execute_generation",
      "validate_output"
    ],
    "enable_self_correction": true,
    "pass_state_via_xml": true
  }
}

enabled

• Type: boolean
• Default: true
• Description: Enable chain prompting workflow
• Impact: Decomposes execution into focused stages
• Recommendation: Keep enabled for quality and traceability

stages

• Type: array of strings
• Default: 7 standard stages (see above)
• Description: Ordered list of workflow stages to execute
• Customization: Can add, remove, or reorder stages
• Standard stages:
  • load_config - Load configuration sources
  • validate_config - Validate against schema
  • merge_config - Merge configurations
  • analyze_spec - Analyze specification
  • plan_execution - Plan agent deployment
  • execute_generation - Run generation
  • validate_output - Validate results
• Optional stages:
  • review_output - Review generated content
  • refine_output - Refine based on review
  • research_existing - Research existing iterations
  • analyze_results - Analyze generation results

enable_self_correction

• Type: boolean
• Default: true
• Description: Enable self-correction loops when validation fails
• Impact: Improves quality through automatic retry and refinement
• Use when: Quality is priority, have retry budget

pass_state_via_xml

• Type: boolean
• Default: true
• Description: Pass state between stages using XML tags
• Impact: Provides clear state visibility and traceability
• Recommendation: Keep enabled for transparency

features

Feature flags for optional capabilities.

{
  "features": {
    "enable_url_strategy": true,
    "enable_theme_evolution": true,
    "enable_cross_iteration_learning": false,
    "enable_automatic_indexing": false
  }
}

enable_url_strategy

• Type: boolean
• Default: true
• Description: Support URL strategy JSON files for progressive web learning
• Use when: Using web enhancement with structured URL progression

enable_theme_evolution

• Type: boolean
• Default: true
• Description: Evolve themes and concepts across iterations
• Impact: Later iterations build on earlier themes
• Use when: Want thematic progression

enable_cross_iteration_learning

• Type: boolean
• Default: false
• Description: Enable learning from previous iterations' outputs
• Impact:
  • Improves quality through learning
  • Increases context usage
  • Slower execution
• Use when: Quality priority, have context budget

enable_automatic_indexing

• Type: boolean
• Default: false
• Description: Auto-generate index.html after each wave
• Impact: Creates navigation for generated files
• Use when: Want easy browsing of outputs

limits

Safety limits to prevent runaway execution.

{
  "limits": {
    "max_iterations": 100,
    "max_file_size_kb": 500,
    "max_total_output_mb": 50,
    "warn_at_iteration": 50
  }
}

max_iterations

• Type: integer
• Minimum: 1
• Default: 100
• Description: Maximum iterations before automatic stop
• Impact: Prevents infinite runaway in infinite mode
• Recommendations:
  • Testing: 10-20
  • Development: 20-50
  • Production: 100-1000
  • Research: 50-100

max_file_size_kb

• Type: integer
• Minimum: 10
• Default: 500
• Description: Maximum size per output file (KB)
• Impact: Catches unexpectedly large files
• Recommendations:
  • Simple HTML: 100-300 KB
  • Complex visualizations: 300-500 KB
  • Large datasets: 500-1000 KB

max_total_output_mb

• Type: integer
• Minimum: 1
• Default: 50
• Description: Maximum total output directory size (MB)
• Impact: Prevents disk space exhaustion
• Recommendations:
  • Testing: 10 MB
  • Development: 20-50 MB
  • Production: 50-200 MB

warn_at_iteration

• Type: integer
• Minimum: 1
• Default: 50
• Description: Show warning to user at this iteration number
• Impact: Prompts user to check progress
• Use when: Long-running batches, want checkpoints

Profiles

Development Profile

File: .claude/config/profiles/development.json

Purpose: Testing, debugging, learning

Key Settings:

• Small batches (3)
• Fewer agents (2)
• Verbose logging
• Review stage enabled
• Lower uniqueness (0.7)
• Conservative limits

Use Cases:

• Testing new specifications
• Debugging issues
• Learning system behavior
• Validating configurations

Trade-offs:

• Slower execution
• More detailed output
• Better for understanding
• Not efficient for scale

Production Profile

File: .claude/config/profiles/production.json

Purpose: Scale, efficiency, throughput

Key Settings:

• Large batches (10)
• More agents (5)
• Minimal logging (warn)
• Review stage disabled
• High uniqueness (0.9)
• Generous limits (1000 iterations)

Use Cases:

• Production deployments
• Large batch generation
• Performance-critical tasks
• Well-tested specifications

Trade-offs:

• Less visibility
• Faster execution
• Assumes specs are correct
• May miss edge cases

Research Profile

File: .claude/config/profiles/research.json

Purpose: Quality, exploration, experimentation

Key Settings:

• Medium batches (5)
• Moderate agents (3)
• Maximum logging (debug, verbose)
• Review stage enabled
• Very high uniqueness (0.95)
• Extended chain prompting (11 stages)
• Extensive web priming (8 URLs)

Use Cases:

• Exploring new domains
• Quality-critical work
• Understanding techniques
• Experimentation

Trade-offs:

• Slower execution
• Higher context usage
• Maximum quality
• Comprehensive analysis

Chain Prompting Configuration

Standard Workflow (7 Stages)

{
  "chain_prompting": {
    "stages": [
      "load_config",
      "validate_config",
      "merge_config",
      "analyze_spec",
      "plan_execution",
      "execute_generation",
      "validate_output"
    ]
  }
}

Execution Flow:

1. Load all config sources
2. Validate against schema
3. Merge hierarchically
4. Analyze specification
5. Plan agent deployment
6. Execute generation
7. Validate outputs

When to Use: Default for most use cases

Extended Workflow (With Review)

{
  "chain_prompting": {
    "stages": [
      "load_config",
      "validate_config",
      "merge_config",
      "analyze_spec",
      "plan_execution",
      "execute_generation",
      "review_output",        // Added
      "validate_output"
    ]
  }
}

Adds: Review stage between generation and validation

When to Use: When quality > speed, have time/context budget

Research Workflow (11 Stages)

{
  "chain_prompting": {
    "stages": [
      "load_config",
      "validate_config",
      "merge_config",
      "analyze_spec",
      "research_existing",    // Added
      "plan_execution",
      "execute_generation",
      "review_output",        // Added
      "refine_output",        // Added
      "validate_output",
      "analyze_results"       // Added
    ]
  }
}

Adds: Research, review, refinement, analysis stages

When to Use: Maximum quality, understanding, experimentation

Minimal Workflow (Speed)

{
  "chain_prompting": {
    "stages": [
      "load_config",
      "merge_config",        // Skip validation
      "analyze_spec",
      "execute_generation"   // Skip planning, validation
    ]
  }
}

Removes: Validation, planning, output validation stages

When to Use: Trusted configs, maximum speed, disposable outputs

Warning: Less safe, may produce errors

Validation

Configuration validation happens in three stages:

1. Schema Validation

Validates against .claude/config/schema.json:

• Type checking (string, integer, boolean, etc.)
• Required fields present
• Value constraints (min/max, enums)
• String patterns (regex)

Example Errors:

ERROR: orchestration.max_parallel_agents
  Expected: integer between 1 and 10
  Actual: 15
  Fix: Reduce to 10 or below

2. Semantic Validation

Validates logical consistency:

• batch_size <= max_iterations
• agent_timeout_ms > web_fetch_timeout_ms
• context_budget_per_agent reasonable
• Output paths valid

Example Warnings:

WARNING: orchestration.max_parallel_agents = 10
  Impact: Very high parallelism may cause performance issues
  Recommendation: Consider 5-8 for better stability

3. Cross-Field Validation

Validates relationships between sections:

• Orchestration vs Quality (review stage vs timeout)
• Orchestration vs Limits (wave size vs max iterations)
• Web Enhancement vs Orchestration (URLs vs timeout)
• Logging vs Performance (verbose + large batch)

Example Issues:

WARNING: Cross-field issue
  Fields: orchestration.max_parallel_agents (5)
          orchestration.infinite_mode_wave_size (20)
          limits.max_iterations (50)
  Issue: May exceed max_iterations in 3 waves
  Recommendation: Reduce wave_size to 10

Validation Commands

Validate Configuration:

/project:validate-config [config_path] [profile]

Examples:

# Validate defaults
/project:validate-config

# Validate profile
/project:validate-config development

# Validate custom config
/project:validate-config examples/custom_config.json

# Validate merged config (development + custom)
/project:validate-config examples/custom_config.json development

Examples

Example 1: Simple Override

Start with defaults, change one parameter:

/project:infinite-config specs/example_spec.md output 5 '{"orchestration":{"max_parallel_agents":6}}'

Result: Uses defaults except max_parallel_agents = 6

Example 2: Profile + Override

Use production profile, add verbose logging:

/project:infinite-config specs/example_spec.md output 20 production '{"logging":{"verbose":true}}'

Result: Production settings + verbose logging

Example 3: Custom Config File

Create my_config.json:

{
  "orchestration": {
    "max_parallel_agents": 4,
    "batch_size": 8
  },
  "quality": {
    "min_uniqueness_threshold": 0.88,
    "enable_review_stage": true
  },
  "web_enhancement": {
    "enabled": true,
    "initial_priming_urls": 5
  }
}

Use it:

/project:infinite-config specs/example_spec.md output 20 custom my_config.json

Example 4: Development → Production

Test with development:

/project:infinite-config specs/example_spec.md test 5 development

Verify quality, then scale with production:

/project:infinite-config specs/example_spec.md prod 100 production

Example 5: Speed Optimization

Create speed config:

{
  "orchestration": {
    "max_parallel_agents": 10,
    "batch_size": 50
  },
  "quality": {
    "enable_validation": false,
    "enable_review_stage": false
  },
  "logging": {
    "level": "error"
  },
  "web_enhancement": {
    "enabled": false
  }
}

Or use command:

/project:configure optimize speed

Example 6: Quality Optimization

Create quality config:

{
  "orchestration": {
    "max_parallel_agents": 2
  },
  "quality": {
    "min_uniqueness_threshold": 0.95,
    "enable_validation": true,
    "enable_review_stage": true,
    "max_retry_attempts": 5
  },
  "web_enhancement": {
    "enabled": true,
    "initial_priming_urls": 8,
    "urls_per_iteration": 3
  },
  "logging": {
    "level": "debug",
    "verbose": true
  },
  "chain_prompting": {
    "stages": [
      "load_config",
      "validate_config",
      "merge_config",
      "analyze_spec",
      "research_existing",
      "plan_execution",
      "execute_generation",
      "review_output",
      "refine_output",
      "validate_output"
    ]
  }
}

Or use command:

/project:configure optimize quality

Best Practices

1. Start with Profiles

Don't create configs from scratch. Use profiles as base:

• Development for testing
• Production for scale
• Research for quality

Then customize with overrides or custom files.

2. Validate Before Running

Always validate configurations:

/project:validate-config my_config.json

Fix errors before executing generations.

3. Test Small, Scale Up

Test with small batches first:

# Test with 3 iterations
/project:infinite-config specs/spec.md test 3 my_config.json

# If good, scale up
/project:infinite-config specs/spec.md prod 50 my_config.json

4. Use Inline Overrides for One-Off Changes

Don't create new config file for single parameter:

# Good: inline override
/project:infinite-config specs/spec.md output 5 production '{"logging":{"verbose":true}}'

# Less good: new config file for one change

5. Document Custom Configs

Add comments (not in JSON, but in adjacent README):

my_config.json - Custom config for D3 visualizations
  - 4 parallel agents for balance
  - 0.88 uniqueness for moderate diversity
  - Review stage enabled for quality
  - 5 priming URLs for D3 techniques

6. Version Control Configs

Keep configurations in version control:

git add .claude/config/profiles/
git add examples/custom_configs/
git commit -m "Add custom D3 visualization config"

7. Compare Configs Before Changing

When modifying configs, compare first:

/project:configure compare current.json proposed.json

Understand differences before switching.

8. Monitor Resource Usage

Watch these metrics:

• Context budget consumption
• Output directory size
• Execution time
• Success/failure rate

Adjust config if issues arise.

9. Profile by Use Case

Create profiles for common scenarios:

• d3_viz_config.json - D3 visualizations
• network_graph_config.json - Network graphs
• quick_test_config.json - Fast testing
• max_quality_config.json - Quality-critical

10. Set Appropriate Limits

Configure limits based on use case:

Testing:

• max_iterations: 10-20
• max_total_output_mb: 10 MB
• warn_at_iteration: 5

Production:

• max_iterations: 100-1000
• max_total_output_mb: 50-100 MB
• warn_at_iteration: 100

Research:

• max_iterations: 50-100
• max_total_output_mb: 50 MB
• warn_at_iteration: 25

Troubleshooting

Problem: Configuration validation fails

Symptoms: Errors when running /project:validate-config

Solutions:

1. Check error messages for specific field issues
2. Compare to working profile configurations
3. Verify values against schema.json
4. Use /project:configure edit to fix interactively

Example:

ERROR: orchestration.max_parallel_agents: 15 exceeds maximum (10)
Fix: Change to 10 or below

Problem: Execution too slow

Symptoms: Generations take too long

Solutions:

1. Increase max_parallel_agents (up to 10)
2. Disable enable_review_stage
3. Reduce logging.level to warn or error
4. Disable web_enhancement if not needed
5. Increase batch_size

Speed Config:

{
  "orchestration": {"max_parallel_agents": 10, "batch_size": 20},
  "quality": {"enable_review_stage": false},
  "logging": {"level": "warn"},
  "web_enhancement": {"enabled": false}
}

Problem: Outputs too similar

Symptoms: Iterations not unique enough

Solutions:

1. Increase quality.min_uniqueness_threshold
2. Enable features.enable_theme_evolution
3. Check spec for diversity requirements
4. Review existing iterations for patterns

Example:

{
  "quality": {
    "min_uniqueness_threshold": 0.9
  },
  "features": {
    "enable_theme_evolution": true
  }
}

Problem: Web fetches failing

Symptoms: Web enhancement not working

Solutions:

1. Increase web_enhancement.web_fetch_timeout_ms
2. Enable web_enhancement.enable_web_search_fallback
3. Reduce web_enhancement.urls_per_iteration
4. Enable web_enhancement.cache_web_content
5. Check URLs are accessible

Example:

{
  "web_enhancement": {
    "web_fetch_timeout_ms": 45000,
    "enable_web_search_fallback": true,
    "cache_web_content": true
  }
}

Problem: Context budget exhaustion

Symptoms: Agents running out of context

Solutions:

1. Increase orchestration.context_budget_per_agent
2. Reduce web_enhancement.initial_priming_urls
3. Reduce web_enhancement.urls_per_iteration
4. Disable features.enable_cross_iteration_learning
5. Reduce number of chain prompting stages

Example:

{
  "orchestration": {
    "context_budget_per_agent": 100000
  },
  "web_enhancement": {
    "initial_priming_urls": 2,
    "urls_per_iteration": 1
  }
}

Problem: Quality issues

Symptoms: Outputs don't meet standards

Solutions:

1. Enable quality.enable_review_stage
2. Increase quality.min_uniqueness_threshold
3. Enable quality.require_spec_compliance_check
4. Increase quality.max_retry_attempts
5. Use research profile
6. Add review/refine stages to chain prompting

Quality Config:

{
  "quality": {
    "min_uniqueness_threshold": 0.95,
    "enable_validation": true,
    "enable_review_stage": true,
    "max_retry_attempts": 5
  },
  "chain_prompting": {
    "stages": [
      "load_config", "validate_config", "merge_config",
      "analyze_spec", "plan_execution", "execute_generation",
      "review_output", "refine_output", "validate_output"
    ]
  }
}

Problem: File size issues

Symptoms: Files too large or hitting limits

Solutions:

1. Adjust limits.max_file_size_kb
2. Check spec requirements
3. Review generated files
4. Modify spec to constrain size

Example:

{
  "limits": {
    "max_file_size_kb": 1000,
    "max_total_output_mb": 100
  }
}

Advanced Topics

Custom Chain Prompting Stages

Add custom stages to workflow:

{
  "chain_prompting": {
    "stages": [
      "load_config",
      "validate_config",
      "merge_config",
      "analyze_spec",
      "research_domain",        // Custom: research domain knowledge
      "plan_execution",
      "execute_generation",
      "peer_review",            // Custom: peer review between agents
      "validate_output",
      "generate_report"         // Custom: generate analysis report
    ]
  }
}

Progressive Sophistication Curves

Configure sophistication progression:

{
  "orchestration": {
    "enable_progressive_sophistication": true,
    "infinite_mode_wave_size": 10
  },
  "progressive_curves": {
    "wave_1": "basic",       // Iterations 1-10
    "wave_2": "intermediate", // Iterations 11-20
    "wave_3": "advanced",    // Iterations 21-30
    "wave_4": "expert"       // Iterations 31+
  }
}

Dynamic Configuration Adjustment

Adjust configuration based on execution metrics:

{
  "adaptive": {
    "enabled": true,
    "adjust_parallel_agents_based_on_success_rate": true,
    "reduce_batch_size_on_timeout": true,
    "increase_retries_on_quality_issues": true
  }
}

Multi-Spec Orchestration

Run multiple specs with shared config:

{
  "multi_spec": {
    "enabled": true,
    "specs": [
      "specs/d3_viz.md",
      "specs/network_graph.md",
      "specs/timeline.md"
    ],
    "output_directories": [
      "output/d3",
      "output/network",
      "output/timeline"
    ]
  }
}

Conclusion

The configuration system provides complete control over infinite loop orchestration:

1. Use profiles as starting points
2. Validate configurations before running
3. Test small before scaling up
4. Monitor resources during execution
5. Document custom configs for reproducibility
6. Version control configurations
7. Optimize for your specific use case

For more information:

• See README.md for overview
• See CLAUDE.md for usage instructions
• See command documentation in .claude/commands/
• See schema reference in .claude/config/schema.json