LeanREA
aicodeking/leanreaReviewed by AI
What it does
leanrea is a single 8 KB instruction file (a "skill") for AI coding assistants such as Claude Code, Codex, Cursor or Gemini CLI. Instead of adding a server and hundreds of extra tools, it tells the assistant which ordinary command to use for each kind of file (Mac and Linux binaries, Electron apps, Android APKs, Go and .NET programs, WASM, SQLite, network captures) and warns it about common traps. The assistant then runs those commands itself. You install it with one npx command, and you supply the inspection tools and a paid AI assistant.
How easy it is to install
C · install 1 or 2 tools · macOS
- macOS C
- Linux C
On macOS, the system the author tested on, it is a C: `npx skills add aicodeking/leanrea` needs Node and a terminal, and the skill only pays off once the command-line inspection tools it routes to are present. Linux is the same grade for the Linux-side tools, but the author's examples lean on macOS commands such as otool and dyld_info. Windows is not covered: the manual install and the commands assume a Unix shell.
Numbers on GitHub
- Stars
- 3,799on Oct 31, 2026
- Downloads
- —all releases
- Growth
- +809stars in 5 days
Stars over time
Facts
- Runs on
- macOS, Linux
- You need to install
- Node.js, Other tools
- Disk space
- not determined
- Accounts or keys
- AI coding assistant (Claude Code, Codex, Cursor, Gemini CLI and similar)
- Cost
- Free, API costs extra · The skill is free (MIT), but it only does anything inside a paid AI coding assistant. The author's own benchmark run of 36 tasks cost about 8 US dollars in model usage.
- License
- MIT (Permissive)
- Can replace
- IDA Pro
- Last update
- Oct 17, 2026
Risk signals (4)
- Attentionreverse engineering may be restricted
The project's whole purpose is reverse engineering: the description says it "replaces REA (Reverse Engineer Anything)" and the benchmark tasks trace features inside "a real Electron app's app.asar" and a desktop app's native add-on. Taking apart software you do not own can break its licence terms and is restricted by law in some countries. Unlike REA's README, this one carries no disclaimer about lawful use.
AI review
- Infothe speed numbers come from the author
The headline "38% faster, 35% cheaper" comes from the author's own benchmark against the project it compares itself to, and his Caveats say "Only 3 reps per cell, one model, one machine" and that Ghidra, Hopper and jadx were not installed. He also notes the gain over using no skill at all is smaller, "about 19% faster".
AI review
- Infothe assistant runs shell commands on your machine
The comparison table says "Tools added to the agent: 0 (uses Bash)", so the skill works by having the assistant run commands directly with your permissions, including "runtime tracing" of the program being inspected.
AI review
- Infovery new and idle since launch
Repository created 2026-10-17 with 3233 stars already, no releases, and the last commit was 2026-10-17 (10 days before this check) with only .gitignore, LICENSE, README.md, bench/ and leanrea/ in it.
AI review
Evidence is shown in its original language.
What Radar could not check
- Whether the benchmark reproduces on another machine, model or app version; the author says the answer keys only match the versions he used.
- Which inspection tools you have to install yourself (asar, jadx, apktool, Ghidra, tshark, binwalk and others are not included).
- Whether the routing commands work on Linux and in agents other than Claude Code.
- What the skill actually tells the assistant to do, since the SKILL.md text is not shown in the README.