vltansky

Skillkill — Security skill for Claude Code

Security community

Audit and undoably clean up stale local agent skills.

How to install Skillkill

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

What Skillkill does

Audit and undoably clean up stale local agent skills.

Alternatives in Security

Portaljs Check Data Quality — Audit a local or remote tabular file (CSV/TSV) for common data quality issues 2.3k ★
Octoboss Clean Contexts — You are the Octoboss — a cross-tentacle orchestrator 1.4k ★
Loongsuite Pilot — Local-first telemetry collector for AI coding agents — unified OpenTelemetry events for Claude Code, Codex, Cu 150 ★

README

skillkill

![skillkill token-cost receipt logo](assets/skillkill-token-receipt.png)

**`npkill` for agent skills.**

Find stale or never-used skill installs and clean them up through an undoable quarantine.

The design goal is boring safety: preview first, explicit cleanup, and a restore path if something was moved by mistake.

Interactive Review

Quick Start

Run without installing:

npx skillkill

Or install globally:

npm install --global skillkill
skillkill

In a real terminal, `skillkill` opens an interactive review. When output is piped, or when `--no-interactive`, `--json`, `--csv`, or `--snapshot` is used, it prints static output instead.

Interactive screens use terminal colors when supported. Set `NO_COLOR=1` to disable color.

What It Does

`skillkill` scans installed skill directories, then checks local agent history for evidence that each skill was actually used.

Default skill roots: ~/.agents/skills, ~/.claude/skills, ~/.codex/skills, and ~/.cursor/skills
Usage: native skill-invocation records and structured skill actions from supported tools
Mention: raw SKILL.md path/name mentions in supported local stores
Cleanup candidates: stale usage or never-used skills past the age threshold
Protected rows: dot-prefixed system skills, recent usage, recent mentions, and omitted skills

Mentions do not prove a skill was invoked, but recent mentions keep a skill out of cleanup candidates because they may indicate use in tools without native attribution.

Safety Model

Default runs do not move files.
Interactive cleanup requires selecting rows, pressing enter to review, then pressing enter again to move them to quarantine.
From the review screen, pressing d, typ