Prusik — AI skill for Claude Code
Open-source safety layer for AI coding agents in production.
How to install Prusik
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open momoElbn/prusik and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Prusik does
Open-source safety layer for AI coding agents in production. Prusik catches destructive actions with an automatic snapshot, a human approval on your phone and one-command undo. Works with Claude Code, Codex, Cursor and OpenCode. Self-hosted, scales to zero.
Alternatives in AI
- Omniget — Udemy & Hotmart course downloader, YouTube downloader (yt-dlp GUI, 1,800+ sites) + desktop app for AI agents 13.9k ★
- Paritok 4b V1 — Non-destructive compression gateway for AI coding agents 1.4k ★
- SoloDawn — Catches and repairs AI hallucinations before delivery: three quality gates, 31 built-in rules, a self-healing 312 ★
README
Slides when you climb, grips when you fall.
**The open-source safety layer that lets AI coding agents work in production,
and catches every dangerous action with a snapshot, a human approval and an undo.**
[](LICENSE) [](#roadmap) [](#works-with) [](#contributing)
[How it works](#how-it-works) · [Features](#features) · [Works with](#works-with) · [Policies](#policies) · [Roadmap](#roadmap) · [Contributing](#contributing)
Why Prusik?
AI coding agents now get the keys to real systems: the production database, the cloud account, the storage buckets. Most of the time that's fine. But once in a while an agent misreads a task and does something that can't be undone. In April 2026, an agent [deleted a company's database and its backups in 9 seconds](https://oecd.ai/en/incidents/2026-04-27-6153).
Prusik sits between the agent and production. Like the climbing knot it's named after, it **slides** freely while the agent works and **grips** the moment something dangerous happens.
- Reads pass straight through. No slowdown for normal work.
- Writes are snapshotted first, and the snapshot is stored where the agent can't reach it.
- Destructive actions wait for a human "yes" on your phone, with a plain-English explanation.
- Anything can be undone with
prusik undo.
How it works
flowchart LR
A[AI codin
Related Skills
Aegis Shellguard
Aegis — the shell guard for AI coding agents. Destructive commands from Claude Code, Codex, or Cursor pause fo
Yea
HTTP was built for browsers. YEA is built for agents. YEA (Your Explicit Approval) is an open protocol for AI
Agentic AI Human Approve Or Reject Agent
Agentic AI agent with Human-in-the-Loop approval using LangGraph and Claude. The agent pauses for human approv
Ethos Gate
A PreToolUse safety gate for autonomous AI agents: blocks irreversible or outward-facing actions (sends, spend
Trackline
An alignment layer for AI agents. Watches whether an agent's actions still match the task, rules and evidence
Mq Sentinel
Production-grade, read-only IBM MQ diagnostic MCP server for AI agents (zero hallucinations). 3-layer safety,
Related Agents
Approval Gate
HumanLayer-inspired approval gate. Enforces human-in-the-loop for HIGH-stakes operations. Classifies actions b
Loom Champion
Loom Champion - Human avatar that promotes quality issues to approved status AND auto-merges Judge-approved PR
Git Snapshots
Git and Snapshot system - full git abstraction plus automatic versioned snapshots. Use proactively when the us