Skip to content

Latest commit

 

History

History
51 lines (41 loc) · 2.1 KB

File metadata and controls

51 lines (41 loc) · 2.1 KB
disclaimer
notice generated_by date
No information within this document should be taken for granted. Any statement or premise not backed by a real logical definition or verifiable reference may be invalid, erroneous, or a hallucination.
GPT-5 Codex
2026-05-22

PURPOSE.md - codebase-mapper

Why This Project Exists

codebase-mapper exists to turn source repositories into inspectable, queryable knowledge bundles. Its core job is to preserve code structure, relationships, provenance, and derived analysis in a form that humans, automation, frontends, and MCP clients can inspect consistently.

The project is built around a verification-first premise: generated facts, LLM-authored annotations, and inferred graph edges are useful only when their source, scope, and failure modes are explicit. Bundle outputs should therefore separate mechanically derived data from advisory or stochastic enrichment, and they should expose enough metadata for downstream consumers to audit claims.

What It Does

  • Maps repositories into RDF and JSON sidecars with file, import, dependency, test, AST, chunk, concept, xref, and optional LLM-enrichment data.
  • Provides CLI entry points for producing bundles from local paths and cloneable Git URLs.
  • Provides a FastAPI backend, React UI, and MCP server for exploring generated bundles.
  • Ships verifiers that keep documentation, schemas, emitted artifacts, and plugin contracts aligned with the live codebase.

Who It Is For

This repository is for engineers and agents that need a grounded view of a codebase before acting on it: refactoring tools, code reviewers, documentation auditors, architecture inspectors, and MCP clients that need read-only repository intelligence.

Non-Goals

  • It is not a replacement for source control or a byte-for-byte repository archive unless a run explicitly emits and retains blobs.
  • It is not a source of unquestioned truth about code intent; inferred and LLM-authored outputs remain derived data.
  • It is not a write-capable remote code execution service. The exposed MCP surface is intentionally read-only.