Skip to content

RLASAF12/agent-undo

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 

Repository files navigation

AgentUndo 🔄

Commit/rollback safety net for irreversible AI agent actions.

Every agentic system eventually does something that can't be undone — sends an email to 500 customers, deletes production rows, drops a DB column. AgentUndo intercepts those actions and gives you a countdown window to commit or roll back before the irreversible happens.


🧪 Live Demo

👉 rlasaf12.github.io/agent-undo


What It Does

Feature Detail
Action interception Queues agent actions before they execute
Risk classification Low / Medium / High based on reversibility
Countdown timers SVG ring timer — auto-commits if you don't act
Rollback prevention Blocks the action and logs exactly what was prevented
Transaction log Full audit trail with timestamps
Pre-scripted scenarios Email blast, DELETE rows, ALTER TABLE, Slack post, S3 delete, webhooks

🎮 Scenarios

📧 Email blast → 500 recipients         (HIGH · 120s timer)
🗑️ DELETE 2,412 DB rows                 (HIGH · 120s timer)
🧬 ALTER TABLE in production            (HIGH · 120s timer)
💬 Post to #company-all (890 members)   (MED  · 60s timer)
📁 Delete S3 bucket (12 GB)             (MED  · 60s timer)
🔔 Trigger outbound webhook             (LOW  · 30s timer)

📂 What's Inside

agent-undo/
└── index.html     # Complete interactive demo (812 lines, zero dependencies)

Single-file. No build step. No backend. Runs on GitHub Pages.


🚀 Quick Start

git clone https://github.com/RLASAF12/agent-undo
cd agent-undo
open index.html   # or just double-click it

Or fork and deploy to GitHub Pages in one click.


💡 Why This Exists

The conversation about AI agent trust has focused on prevention (don't let the agent do X) and detection (alert when X happens). But there's a gap nobody's filling: the window between "agent decided to act" and "action executed."

AgentUndo sits in that gap. It's not a firewall. It's a commit log.

"The bottleneck in agentic workflows isn't code generation speed — it's verification capacity."
— HN consensus, June 2026


🏗️ Agent Trust Layer Series

This is Build #8 in the series:

# Tool Category
1 AgentGate Prevention
2 AgentSentinel Detection
3 AgentReceipt Accountability
4 agent-breaker Containment
5 pr-flood Containment
6 agent-postmortem Diagnosis
7 AgentDrift Detection
8 AgentUndo Recovery

Built by @RLASAF12 · Security checklist passed 2026-06-27

About

Commit/rollback safety net for irreversible AI agent actions. Agent Trust Layer · Build #8.

Topics

Resources

License

Stars

0 stars

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages