Deadskills


Every installed skill adds its name and description to every prompt you send. That cost is paid on every message, whether the skill gets used or not. `deadskills` reads your local transcripts and shows which skills earn their seat, and which ones are probably safe to remove.

[](https://www.npmjs.com/package/deadskills) [](https://www.npmjs.com/package/deadskills) [](LICENSE) [](https://nodejs.org)

Quick start

**Requires Node 18+.** Auto detects Claude Code (`~/.claude`) and Codex (`~/.codex`) when those directories exist. No config, no login, no telemetry.

npx deadskills

That is the whole workflow. Everything below is reference.


Install

npx deadskills               # one-off run, always latest (recommended)
npm install -g deadskills    # global install: run deadskills anywhere

From source (development):

git clone https://github.com/anandsaini18/deadskills
cd deadskills
make install && make link    # deadskills on PATH from local build

Zero runtime dependencies. The published package is one ~22 KB file you can read in a sitting.

Commands

Command	What it does
deadskills	Full report for all detected agents
deadskills dead	List dead skills only
deadskills doctor	Parse health: are your transcripts being read correctly?
deadskills --since 30d	Limit analysis to a window (30d, 8w, 6m, or an ISO date)
deadskills --json	Machine readable output (schema)