Discover Capabilities

Discover CodeCompass Capabilities

Purpose

This meta-skill teaches how to efficiently discover what CodeCompass can do without manual exploration or rediscovery.

When to Use

• ✅ Session start (understand available capabilities)
• ✅ User asks: "What can CodeCompass do?"
• ✅ Before implementing features (check if exists)
• ✅ Exploring unfamiliar parts of codebase
• ✅ Planning complex workflows (know available tools)

Discovery Layers

Layer 1: Quick Reference (CLAUDE.md)

Purpose: Project overview and entry point Location: /CLAUDE.md Token cost: ~2,000 tokens Content: Tech stack, infrastructure, basic commands

Read for:

• Project background
• Technology choices
• Infrastructure setup
• Quick commands

Layer 2: Structured Discovery (capabilities.json) ⭐

Purpose: Complete module and capability map Location: .ai/capabilities.json Token cost: ~2,500 tokens Content: All domains, modules, relationships, entry points

Read for:

• What domains exist
• What modules are available
• What each module can do
• How modules relate to each other
• When to use which module

Layer 3: Skills Index (this directory)

Purpose: How-to workflows and perspectives Location: .claude/skills/ Token cost: ~700 tokens (metadata scan) Content: Workflows, roles, patterns

Read for:

• How to accomplish complex tasks
• What workflows exist
• Which skill to apply

Layer 4: Semantic Search (Weaviate)

Purpose: Implementation-level discovery Location: Weaviate database Token cost: Variable (query results) Content: Actual code, indexed semantically

Use for:

• Finding specific implementations
• Similar code patterns
• Deep code understanding

Layer 5: Source Code

Purpose: Implementation details Location: src/modules/ Token cost: Variable (file reading)

Read for:

• Exact implementation
• Edge cases
• Internal logic

Discovery Workflow

Quick Discovery (5 seconds)

# 1. Read capabilities.json structure
cat .ai/capabilities.json | jq '.domains | keys'
# Returns: ["code_analysis", "semantic_search", "requirements_extraction", "vault_integration", "system_operations"]

# 2. Check specific domain
cat .ai/capabilities.json | jq '.domains.code_analysis'
# Returns: modules, capabilities, entry_points, when_to_use

Structured Discovery (Recommended)

Step 1: Read capabilities.json

{
  "domains": {
    "code_analysis": {
      "modules": ["ast-analyzer", "yii2-analyzer", ...],
      "capabilities": ["parse_php_ast", "extract_yii2_controllers", ...],
      "entry_points": {...}
    }
  }
}

Step 2: Identify relevant domain

• User wants: "Analyze Yii2 project" → code_analysis domain
• User wants: "Search for payment logic" → semantic_search domain
• User wants: "Extract requirements" → requirements_extraction domain

Step 3: Check module capabilities

{
  "code_analysis": {
    "modules": ["yii2-analyzer"],
    "capabilities": ["extract_yii2_controllers", "extract_yii2_models"],
    "entry_points": {
      "cli": "codecompass analyze:yii2 <path>"
    }
  }
}

Step 4: Load relevant skill

• For Yii2 analysis → Load analyze-yii2-project.md skill
• For requirements → Load extract-requirements.md skill

Semantic Discovery (Deep Dive)

When capabilities.json shows a module exists but you need implementation details:

Use Weaviate semantic search:

pnpm run cli -- search:semantic "how does Yii2 controller extraction work"

Returns:

• Relevant source files
• Similar code patterns
• Implementation examples

Example Flows

Example 1: "Can CodeCompass analyze database schemas?"

Step 1: Check capabilities.json

cat .ai/capabilities.json | jq '.domains.code_analysis.modules[]' | grep database
# Returns: "database-analyzer"

✅ Answer: Yes, via database-analyzer module

Step 2: Check entry points

cat .ai/capabilities.json | jq '.domains.code_analysis.entry_points'

Step 3: Check if skill exists

ls .claude/skills/ | grep database
# If exists, load that skill for workflow
# If not, use capabilities.json guidance directly

Example 2: "How do I search code semantically?"

Step 1: Check capabilities.json

cat .ai/capabilities.json | jq '.domains.semantic_search'

Returns: modules, capabilities, dependencies

Step 2: Load semantic-search skill

cat .claude/skills/semantic-search.md

Follow the workflow in the skill

Example 3: "What modules handle authentication?"

Step 1: Semantic search in Weaviate

pnpm run cli -- search:semantic "authentication and authorization modules"

Step 2: Verify in capabilities.json

cat .ai/capabilities.json | jq '.module_relationships'

Discovery Decision Tree

┌─────────────────────────────────────┐
│ USER REQUEST                        │
└─────────────────────────────────────┘
              ↓
        ┌─────────┐
        │ What?   │ "What can CodeCompass do?"
        └─────────┘
              ↓
    Read: .ai/capabilities.json
              ↓
        ┌─────────┐
        │ How?    │ "How do I use X?"
        └─────────┘
              ↓
    Load: .claude/skills/<relevant>.md
              ↓
        ┌─────────┐
        │ Where?  │ "Where is X implemented?"
        └─────────┘
              ↓
    Search: Weaviate semantic search
              ↓
        ┌─────────┐
        │ Details │ "How does X work internally?"
        └─────────┘
              ↓
    Read: src/modules/<module>/<file>

Avoid Rediscovery Pattern

❌ Old Pattern (Inefficient)

1. User asks: "Can CodeCompass do X?"
2. Claude explores with Grep: 1,000+ tokens
3. Claude reads multiple files: 5,000+ tokens
4. Claude builds mental model: slow, error-prone
5. Total: 6,000+ tokens, slow, may miss things

✅ New Pattern (Efficient)

1. User asks: "Can CodeCompass do X?"
2. Claude reads capabilities.json: 2,500 tokens (cached)
3. Claude checks domain → module → capability
4. Claude loads skill if complex workflow needed
5. Total: 2,500-4,000 tokens, instant, accurate

Module Relationship Understanding

capabilities.json includes module_relationships:

{
  "indexing": {
    "depends_on": ["vectorizer", "weaviate"],
    "used_by": ["search", "vault"],
    "provides": "embedding_generation_pipeline"
  }
}

Use this to:

• Understand module dependencies
• Trace data flow
• Plan refactoring
• Debug integration issues

Common Workflows Reference

capabilities.json includes common_workflows:

{
  "analyze_legacy_yii2": [
    "1. Verify infrastructure: codecompass health",
    "2. Analyze project: codecompass analyze:yii2 <path>",
    ...
  ]
}

Use as:

• Quick reference for common tasks
• Starting point for complex workflows
• Checklist for multi-step processes

Maintenance Notes

This skill references:

• .ai/capabilities.json - Keep this file updated when modules change
• .claude/skills/ - Skills directory for workflows
• CLAUDE.md - Project entry point

Update triggers:

• New module added → Update capabilities.json
• Module removed → Update capabilities.json
• New workflow discovered → Consider creating new skill

Related Skills

• analyze-yii2-project.md - Framework-specific workflow after discovering Yii2 modules
• semantic-search.md - How to search capabilities after discovery
• extract-requirements.md - What to do after discovering requirements modules

Related Modules

From .ai/capabilities.json:

• All 5 domains: code_analysis, semantic_search, requirements_extraction, vault_integration, system_operations
• All 21+ modules documented in capabilities.json

Remember: capabilities.json is the single source of truth for "WHAT exists". Skills teach "HOW to use it".