PruvaGraph: Instructions file for Claude Code

PruvaGraph — Codebase Knowledge Graph

This project uses PRUVALEX PruvaGraph to maintain a pre-built knowledge graph (pruvagraph-out/graph.json). This graph encodes the full structure of the codebase at a fraction of the token cost of reading raw files.

---

MANDATORY: Use Graph Tools Before Reading Files

Before opening any source file, reading any directory listing, or using grep for any question about codebase structure, call one of the MCP tools below.

The graph already contains: imports, dependencies, callers, community clusters, type signatures, and architectural summaries — fully offline, no API cost.

Reading raw files when the graph has the answer wastes 5×–71× more tokens.

If you want to know…	Use this tool FIRST
How module X connects to Y	query_graph "how does X connect to Y?"
What a class/function does	get_summary "ClassName"
What file A imports	get_dependencies "path/to/file.py"
Who calls function F	find_callers "function_name"
Architectural overview	list_communities
What changed structurally	get_graph_diff
What breaks if X changes	analyze_impact "SymbolName"
Monorepo packages	list_packages
Cost savings so far	cost_report

Only open a raw file if:

1. The graph tool returns "not found" for the specific symbol, OR
2. You need to edit the file (structure info still comes from graph first), OR
3. The question is about the content of a specific function body (not its connections).

---

Rebuild the Graph

pruvagraph .            # Full rebuild
pruvagraph . --update   # Incremental (changed files only)
pruvagraph . --dry-run  # Estimate cost first
pruvagraph . --force    # Force full rebuild, ignore cache

Output Files

• pruvagraph-out/graph.json — Full knowledge graph (MCP server reads this)
• pruvagraph-out/graph.html — Interactive visualiser (open in browser)
• pruvagraph-out/GRAPH_REPORT.md — Architectural summary + god nodes
• pruvagraph-out/cost_report.json — Cost analytics + query benchmarks