anandsaini18/deadskills — explained in plain English
Analysis updated 2026-05-18
Find out which installed AI-agent skills are never actually invoked and safe to remove.
See an estimated token cost that installed skills add to every prompt you send.
Spot skills that used to be useful but have gone quiet for 90 or more days.
Export a machine-readable JSON report to build your own dashboard on top of.
| anandsaini18/deadskills | karminski/codevinci | ningsiii/clickdeck | |
|---|---|---|---|
| Stars | 35 | 35 | 35 |
| Language | TypeScript | TypeScript | TypeScript |
| Setup difficulty | easy | moderate | easy |
| Complexity | 1/5 | 3/5 | 2/5 |
| Audience | developer | developer | vibe coder |
Figures from each repo's GitHub metadata at analysis time.
Requires Node 18 or newer, auto-detects Claude Code and Codex data folders with zero configuration.
deadskills is a command line tool that looks at your local usage history with AI coding assistants like Claude Code or Codex and tells you which of your installed skills are actually being used. A skill is a small add-on that teaches the assistant a specific task, but every installed skill adds its name and description to every single prompt you send, whether it ends up being used or not. This tool helps you see that hidden cost and decide what to remove. You run it with a single command, npx deadskills, and it automatically finds your Claude Code and Codex data folders if they exist, with no login, configuration, or network calls involved, since everything runs and stays on your own machine. The report it produces shows a single number for how many tokens, the units language models are billed and measured by, your installed skills are adding to every prompt, alongside how often each skill was actually invoked based on your own transcript history, not guesswork. Each skill gets sorted into one of three states. An active skill has been invoked at least once and is earning its place. A zombie skill was used before but has gone quiet for over 90 days. A dead skill was installed but has never been invoked a single time. The report lists these with their estimated per-prompt cost so you can see exactly what removing each one would save. Under the hood, it reads each skill's description file to learn its name and estimated token cost, parses your session transcripts, and combines invocation counts with token estimates to flag which skills are dead or going stale. Token counts are clearly labeled as estimates, since the tool cannot know your exact prompt caching setup, but they are described as directionally accurate. The published package has no runtime dependencies and is small enough to read in one sitting. It currently supports Claude Code and Codex, with support for more agents welcomed through community contributions. The tool is released under the MIT license, and its report data format is released separately under CC0.
A local command-line tool that scans your Claude Code or Codex transcripts to show which installed skills you actually use, and which are dead weight.
Mainly TypeScript. The stack also includes TypeScript, Node.js, Claude Code.
Use freely for any purpose, including commercial use, as long as you keep the copyright notice.
Setup difficulty is rated easy, with roughly 5min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
double-check against the repo, no cap.