7.4 KiB

Raw Blame History

Context Compression Utility

Purpose

Provide lean utility for token optimization and relevance filtering across all BMAD agents. This utility enables efficient context management by reducing token usage while preserving essential information based on agent-specific requirements.

Compression Strategies

Agent-Specific Compression Ratios

Dev Agents: 0.9 compression ratio (aggressive optimization)
Planning Agents: 0.7 compression ratio (balanced optimization)
Hybrid Agents: Adaptive compression based on current mode

Compression Techniques

Token-Level Optimization

Remove redundant whitespace and formatting
Eliminate unnecessary punctuation and filler words
Compress verbose explanations to essential points
Optimize sentence structure for clarity and brevity

Content-Level Compression

Summarize lengthy descriptions and explanations
Remove duplicate or repetitive information
Consolidate related concepts and ideas
Extract key points from verbose content

Structural Optimization

Flatten nested information hierarchies
Combine related sections and topics
Remove empty or low-value content blocks
Optimize markdown structure for efficiency

Compression Procedures

1. Pre-Compression Analysis

Content Assessment LLM: Analyze content structure, token distribution, and compression opportunities

Token Distribution Analysis

Identify high-token content areas
Calculate compression potential by section
Assess information density and value
Determine optimal compression targets

Content Categorization

Content Classification:
- Essential (preserve): {{essential_content_percentage}}%
- Important (compress): {{important_content_percentage}}%
- Supplementary (aggressive compression): {{supplementary_content_percentage}}%
- Removable (eliminate): {{removable_content_percentage}}%

2. Agent-Specific Compression

For Dev Agents (0.9 compression ratio):

Aggressive Technical Compression

Remove non-technical explanations and context
Compress verbose documentation to bullet points
Eliminate planning discussions and business context
Focus on actionable technical information only

Code-Focused Optimization

Preserve code examples and technical specifications
Remove explanatory text around code snippets
Compress error descriptions to essential details
Maintain technical accuracy while reducing verbosity

Implementation-Centric Filtering

Keep task requirements and acceptance criteria
Remove strategic discussions and background context
Preserve debugging information and edge cases
Eliminate stakeholder and business information

For Planning Agents (0.7 compression ratio):

Strategic Context Preservation

Maintain high-level goals and business objectives
Preserve stakeholder requirements and constraints
Keep decision rationale and strategic context
Compress operational details while maintaining strategic view

Collaborative Information Compression

Summarize cross-functional collaboration details
Compress meeting notes to key decisions and outcomes
Preserve important stakeholder communications
Maintain project status and milestone information

Balanced Technical Compression

Compress technical details to strategic implications
Remove low-level implementation specifics
Preserve architecture decisions and technical constraints
Maintain technical context relevant to planning decisions

3. Compression Execution

Structured Compression Process

Identify content blocks for compression
Apply agent-specific compression rules
Validate information preservation
Optimize token efficiency
Verify compression ratio targets

Content Transformation Rules

Compression Transformations:
- Long paragraphs → Bullet points
- Verbose explanations → Key concepts
- Detailed examples → Essential patterns
- Historical context → Current relevance
- Redundant information → Single reference

4. Quality Preservation

Information Integrity Checks

Verify essential information is preserved
Confirm no critical details are lost
Validate relationships between concepts remain intact
Ensure compressed content maintains accuracy

Compression Quality Metrics

Quality Assessment:
- Information Completeness: {{completeness_score}}/10
- Accuracy Preservation: {{accuracy_score}}/10
- Readability Maintenance: {{readability_score}}/10
- Token Efficiency: {{efficiency_score}}/10

Rollback Procedures

Identify compression failures or quality issues
Restore original content when compression degrades quality
Apply alternative compression strategies
Maintain compression quality standards

Utility Functions

Adaptive Compression

compress_for_agent_type(content, agent_type, target_ratio):
  - Apply agent-specific compression rules
  - Target specified compression ratio
  - Preserve agent-relevant information
  - Return optimized content within token limits

Token Optimization

optimize_token_usage(content, token_limit):
  - Analyze current token usage
  - Apply progressive compression techniques
  - Maintain information quality
  - Achieve target token count

Content Summarization

summarize_content(content, summary_ratio):
  - Extract key concepts and ideas
  - Preserve essential information
  - Maintain logical flow and structure
  - Achieve target summary length

Relevance Filtering

filter_by_relevance(content, relevance_threshold, agent_type):
  - Score content relevance for agent type
  - Remove content below threshold
  - Preserve highly relevant information
  - Optimize for agent-specific needs

Output Format

Compression Results

## Content Compression Results

**Original Content:**
- Token Count: {{original_token_count}}
- Content Blocks: {{original_block_count}}
- Word Count: {{original_word_count}}

**Compressed Content:**
- Token Count: {{compressed_token_count}}
- Compression Ratio: {{compression_ratio}}
- Space Saved: {{space_saved_percentage}}%
- Quality Score: {{quality_score}}/10

**Compression Summary:**
- Blocks Compressed: {{compressed_blocks}}
- Blocks Removed: {{removed_blocks}}
- Information Preserved: {{preservation_percentage}}%
- Agent Optimization: {{agent_type}} focused

Compressed Content

{{compressed_content}}

Performance Monitoring

Compression Metrics

Compression ratio achievement vs. target
Token reduction efficiency
Information preservation quality
Processing time and performance impact

Quality Indicators

Content accuracy maintenance
Information completeness scores
User satisfaction with compressed content
Agent performance with compressed context

Optimization Tracking

Compression algorithm effectiveness
Agent-specific compression success rates
Quality degradation incidents
Performance improvement opportunities

Configuration

Agent-Specific Settings

compression_settings:
  dev_agents:
    target_ratio: 0.9
    focus: technical_content
    preserve: code_and_specs
    remove: planning_context
  
  planning_agents:
    target_ratio: 0.7
    focus: strategic_content
    preserve: business_context
    remove: implementation_details
    
  quality_thresholds:
    minimum_accuracy: 0.95
    minimum_completeness: 0.90
    minimum_readability: 0.85

Uses configuration from core-config.yaml context_engineering section for agent-specific compression settings and quality standards.

7.4 KiB Raw Blame History