explore - Claude MCP Skill

# Explore Agent — Code Property Graph First

## Protocol

1. **ALWAYS use codebase-memory-mcp tools FIRST** for any code discovery:
   - `search_graph(name_pattern/label/qn_pattern)` — find functions, classes, routes
   - `trace_path(function_name, mode=calls|data_flow|cross_service)` — trace call chains
   - `get_code_snippet(qualified_name)` — read source code
   - `query_graph(query)` — complex Cypher patterns
   - `get_architecture(aspects)` — project structure overview
   - `search_code(pattern)` — graph-augmented grep

2. **Fall back to Grep/Glob/Read** ONLY for text content search, config values, or non-code files.

3. **If the project is not indexed:** call `index_repository(repo_path)` FIRST, then use the graph tools.

4. **For intent understanding:** use `mcp__codebase-memory-mcp__query_graph` with queries that look for ReasonNodes, contracts, and drift events.

## Why This Beats grep

- grep finds text; the graph finds **definitions** with structural context
- grep returns files; `trace_path` returns **call chains** with data flow
- grep is line-based; `query_graph` can answer "what depends on this?" in one query
- The graph surfaces **intent** (WHY code exists) through ReasonNodes and iCPG contracts

## Speed Comparison

| Task | grep Approach | Graph Approach |
|------|--------------|----------------|
| Find all callers of `authenticate()` | grep for name, filter noise | `trace_path("authenticate", inbound)` — instant |
| What handles payment errors? | grep "payment.*error", read files | `search_graph(query="payment error handler")` |
| Architecture overview | Read 20+ files manually | `get_architecture()` — one call |