---

name: kosmos-xray description: Context-efficient codebase exploration using AST analysis. Use when exploring Kosmos architecture, understanding code structure, or preparing documentation for AI programmers. Triggers: xray, map structure, skeleton, interface, architecture, explore kosmos, warm start, token budget, context compression.

Kosmos X-Ray Skill

Specialized tools for analyzing the Kosmos codebase efficiently within limited context windows. Uses AST parsing to extract structural information (classes, methods, signatures) without loading implementation details, achieving ~95% token reduction.

Enhanced Features (v2)

The skeleton extractor now captures:

• Pydantic/dataclass fields - name: str = Field(...) visible in output
• Decorators - @dataclass, @property, @tool shown above definitions
• Global constants - CONFIG_VAR = "value" at module level
• Line numbers - Every definition includes # L{line} for navigation

IMPORTANT: Always use these features when exploring - they reveal data structures that would otherwise appear as empty pass statements.

When to Use This Skill

• Exploring the codebase - Map directory structure before diving into files
• Understanding architecture - Extract class hierarchies and dependencies
• Understanding data models - Skeleton shows Pydantic fields that define the data
• Onboarding - Generate documentation for new AI programmers
• Context management - Identify large files that should use skeleton view instead of full read

Core Tools

1. mapper.py - Directory Structure Map

Shows file tree with token estimates. Identifies context hazards (large files).

# Map entire project
python .claude/skills/kosmos-xray/scripts/mapper.py

# Map specific directory
python .claude/skills/kosmos-xray/scripts/mapper.py kosmos/workflow/

# Get summary only (no tree) - RECOMMENDED FIRST STEP
python .claude/skills/kosmos-xray/scripts/mapper.py --summary

# JSON output for parsing
python .claude/skills/kosmos-xray/scripts/mapper.py --json

2. skeleton.py - Interface Extraction (Enhanced)

Extracts Python file skeletons via AST. Now shows Pydantic fields, decorators, constants, and line numbers.

# Single file skeleton (includes line numbers by default)
python .claude/skills/kosmos-xray/scripts/skeleton.py kosmos/workflow/research_loop.py

# Directory with pattern filter
python .claude/skills/kosmos-xray/scripts/skeleton.py kosmos/ --pattern "**/base*.py"

# Filter by priority (critical, high, medium, low) - USE THIS FOR ONBOARDING
python .claude/skills/kosmos-xray/scripts/skeleton.py kosmos/ --priority critical

# Include private methods (_method) for internal understanding
python .claude/skills/kosmos-xray/scripts/skeleton.py kosmos/agents/ --private

# Omit line numbers if not needed
python .claude/skills/kosmos-xray/scripts/skeleton.py kosmos/config.py --no-line-numbers

# JSON output for programmatic use
python .claude/skills/kosmos-xray/scripts/skeleton.py kosmos/models/ --json

What skeleton.py reveals:

# Before (old behavior): Data models appeared empty
class Hypothesis(BaseModel):
    pass

# After (enhanced): Full data structure visible
@dataclass
class PaperAnalysis:  # L34
    paper_id: str  # L36
    executive_summary: str  # L37
    confidence_score: float  # L42

3. dependency_graph.py - Import Analysis

Maps import relationships between modules. Identifies architectural layers and circular dependencies.

# Analyze dependencies (text output)
python .claude/skills/kosmos-xray/scripts/dependency_graph.py kosmos/

# With root package name (recommended)
python .claude/skills/kosmos-xray/scripts/dependency_graph.py kosmos/ --root kosmos

# Focus on specific area
python .claude/skills/kosmos-xray/scripts/dependency_graph.py kosmos/ --focus workflow

# Generate Mermaid diagram for documentation - USE FOR WARM_START.md
python .claude/skills/kosmos-xray/scripts/dependency_graph.py kosmos/ --root kosmos --mermaid

# Combined: Mermaid focused on workflow
python .claude/skills/kosmos-xray/scripts/dependency_graph.py kosmos/ --root kosmos --mermaid --focus workflow

# JSON output
python .claude/skills/kosmos-xray/scripts/dependency_graph.py kosmos/ --json

Recommended Workflow (Use ALL Features)

1. Survey first - mapper.py --summary to see codebase size and large files
2. X-ray critical classes - skeleton.py --priority critical to see core interfaces WITH FIELDS
3. Generate architecture diagram - dependency_graph.py --mermaid for visual map
4. Verify imports - Run import checks before documenting entry points
5. Read selectively - Only read full implementation when skeleton isn't enough

Best Practices

DO:

• Always use --priority critical first to understand core architecture
• Use --mermaid output for documentation diagrams
• Check line numbers when you need to reference specific code
• Use --private when understanding internal agent behavior
• Verify imports before documenting them as entry points

DON'T:

• Read full files when skeleton would suffice (wastes context)
• Ignore large file warnings from mapper.py
• Skip the Pydantic fields - they define the data contracts
• Forget to include line numbers in documentation references

Integration with kosmos_architect Agent

This skill is automatically loaded by the kosmos_architect agent. You can also use it directly for targeted analysis.

# Use the agent for full onboarding documentation (uses ALL features)
@kosmos_architect generate

# Or use individual tools directly
@kosmos-xray Map the workflow directory

Configuration Files

• configs/ignore_patterns.json - Directories and files to skip
• configs/priority_modules.json - Module priority levels and patterns

Context Budget Guidelines

Operation	Typical Tokens	Use When
mapper.py --summary	~500	First exploration
mapper.py full	~2-5K	Understanding structure
skeleton.py (1 file)	~200-500	Understanding interface
skeleton.py --priority critical	~5K	Core architecture
dependency_graph.py text	~2-3K	Architecture analysis
dependency_graph.py --mermaid	~500	Documentation diagrams
Full file read	Varies	Need implementation details

For detailed API documentation, see reference.md. For quick command reference, see CHEATSHEET.md.