Damping 卡在你的 AI 編碼 agent（Claude Code、Cursor、Codex）與真實世界之間：跑破壞性 shell 指令、呼叫高風險 MCP 工具之前，先攔下來對照政策，讓你決定放行或擋下——終端機與 MCP 伺服器共用同一套政策引擎、同一份稽核紀錄，不是只顧其中一邊。每一條內建規則都對應一起真實發生過的事故：agent 誤刪使用者整個家目錄、Replit agent 刪掉正式資料庫波及上千位主管、Claude Code 對正式 AWS 帳號跑 terraform destroy 差點清空兩年半的課程資料、惡意指令藏在 CLAUDE.md 裡誘騙 agent 偷 SSH 金鑰與加密貨幣錢包。開源（Apache-2.0 授權），零遙測，一行指令安裝，GitHub 上已有上百次提交、持續在更新。
Damping sits between your AI coding agent (Claude Code, Cursor, Codex) and the real world: before a destructive shell command or a risky MCP tool call runs, it checks against a policy and lets you decide — the terminal and MCP servers share the same policy engine and the same audit trail, not just one or the other. Every built-in rule maps to a real, documented incident: an agent that wiped a user's entire home directory, a Replit agent that deleted a production database affecting over a thousand executives, a Claude Code session that ran terraform destroy on a live AWS account and nearly erased two and a half years of course data, malicious instructions hidden in CLAUDE.md that tricked agents into exfiltrating SSH keys and crypto wallets. Open source (Apache-2.0), zero telemetry, install with one command — over a hundred commits on GitHub and actively maintained.