janghotan

Repo Onboarding — AI skill for Claude Code

AI community

Persistent, commit-tracked codebase architecture and convention mapper for Claude Skills and AI coding agents.

How to install Repo Onboarding

This entry records only its repository, not the path inside it, so there is no exact command to give. Open janghotan/repo-onboarding and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Repo Onboarding does

Persistent, commit-tracked codebase architecture and convention mapper for Claude Skills and AI coding agents.

Alternatives in AI

Lumenx Model Onboarding — LumenX 模型接入与文档更新流程 - 厂商文档抓取、model_catalog 更新、运行时范围判断、端到端验证 1.2k ★
Codebase Digest — 🗜️ Codebase-digest is your AI-friendly codebase packer and analyzer 375 ★
Fallow Skills — Agent skills for fallow, codebase intelligence for TypeScript and JavaScript 116 ★

README

[English](README.md) | [简体中文](README.zh-CN.md)

**Description**: Persistent, commit-tracked codebase architecture and convention mapper for Claude Skills and AI coding agents. **Topics**: `claude-skills` · `ai-agents` · `developer-tools` · `onboarding`

`repo-onboarding` is a Claude Skill (`SKILL.md` + zero-dependency discovery scripts) that gives Claude an accurate, persistent mental map of any codebase it is dropped into. Instead of burning tokens re-scanning directories from scratch every session — or guessing from shallow filename matches — Claude builds a commit-tracked map in `.claude/repo-map/`, updates it incrementally via `git diff ..HEAD --stat`, and answers `"how does X work here?"` or `"where do I change Y?"` with verified `file:line` citations.

Why `repo-onboarding`?

When LLMs onboard to unfamiliar codebases, they usually re-derive structure from scratch every session, wasting tokens and giving shallow or wrong answers:

Shallow Guessing: Answering architectural or modification questions after reading only one or two files, missing custom error hierarchies, auth middleware, or structured logging wrappers.
Repeated Cold-Start Scans: Re-running broad file trees and grep sweeps at the start of every session, wasting context window capacity and latency on unchanged code.

`repo-onboarding` persists and grounds that understanding across sessions:

First Invocation (Full Structural Scan): Runs a language-agnostic discovery pass (scripts/scan.sh) and synthesizes five structured files into .claude/repo-map/.
Subsequent Invocations (Git Commit Diff Check): Compares last_built_commit in .claude/repo-map/index.json against HEAD (scripts/diff-check.sh). If unchanged, the map is used immediately. If commits landed, only the affected directories are re-scanned (`git diff ..HE