SIN-Code Semantic Codebase Knowledge Graphs (SCKG)

A real dependency graph for AI coding agents — Tree-sitter parsing, Git-history intent, and blast-radius impact analysis instead of a dumb vector store.

Part of the SIN-Code agent-engineering stack.

Why

Vector-database RAG retrieves text that looks similar. It does not know that changing parse() will break 14 call sites. SCKG builds an actual MultiDiGraph of symbols, calls, file-containment and Git-commit intents, so an agent can ask structural questions: what calls this? what breaks if I change it? where are the architectural hubs?

Features

• Tree-sitter AST parsing for Python, JavaScript and TypeScript.
• Call / containment edges resolved across the repository.
• Git-history intent extraction — commits classified as refactor / feature / fix / docs and linked to the files they touched.
• Blast-radius impact analysis with a normalized risk score.
• Architecture overview — top hubs by out-degree, total nodes/edges.
• Persistent graph stored as JSON under .sin/.
• CLI (sckg) and MCP server for agent integration.
• Graceful degradation — a missing language grammar is skipped with a warning, never a crash.

Quickstart

pip install -e .
sckg build            # parse the current repo into .sin/knowledge.graph
sckg arch             # show hubs and totals
sckg find parse       # locate a symbol by name
sckg impact "src/x.py:function:parse"   # blast radius

Documentation

• INSTALL.md — installation and verification
• docs/USAGE.md — CLI commands and MCP tools
• docs/CONFIGURATION.md — config.yaml reference
• CONTRIBUTING.md — development workflow
• CHANGELOG.md — release notes

MCP integration

# ~/.config/opencode/config.yaml
mcpServers:
  sckg:
    command: sckg
    args: [serve]

Exposed tools: find_symbol, impact_analysis, architecture_overview, downstream_deps.

License

MIT — see LICENSE.