// observe

Bug Bot

built, not on yet

Someone emails a bug report. Robots investigate, write the fix, and wait for me to say go.

Bug Bot is a daemon (a little program that runs quietly in the background) that watches for bug report emails from trusted senders. When one arrives, it spins up a team of AI assistants to investigate and write a fix. Then it emails me a full report and a draft reply, and waits for me to say "go" before touching production.

  • Python
  • Claude Code
  • systemd
Flat illustration: a friendly robot inspects a code bug and patches a fix, paused at a human approval button before anything ships.
The whole point is that I shouldn't have to drop what I'm doing every time a bug comes in. Bug Bot handles the tedious first two hours automatically: reading the report, digging through the code, writing the fix, even drafting the reply to the person who sent it. But it won't deploy anything until I give the green light, by email or Signal message. It's the right balance: automation for the grunt work, human judgment for the final call.

What it does

When someone trusted emails a bug report to my address, Bug Bot intercepts it before I even see it. It creates a safe sandbox copy of the code, then launches a small team of AI assistants working in sequence, each double-checking the one before. When they're done, Bug Bot emails me a plain-English report of what was found, what was changed, and a draft reply for the person who reported it. The fix sits on a branch (nothing touches the live app) until I reply "GO" to the email or sends a Signal message. Only then does Bug Bot run the deploy. Security-flagged bugs require an extra human review step and can never auto-deploy no matter what.

Why it's neat

Robots do the boring first two hours

Investigation, code digging, drafting a reply - Bug Bot handles all of it so I can focus on things that actually need my brain.

Nothing ships without a human saying so

The daemon holds the deploy key, not the AI; no fix can reach production until I explicitly approve it.

A whole team reviews the fix

Rather than trusting one assistant, Bug Bot runs a few assistants working in sequence, each double-checking the one before.

How it works

1. A bug report email arrives

Signal-router (my email-routing daemon) spots the report and drops it in Bug Bot's queue.

2. A team of assistants gets to work

Bug Bot opens a safe copy of the codebase, then runs a chain of AI assistants - researcher, implementer, reviewer - each building on the last.

3. I get a full report by email

Bug Bot sends a plain-English summary of the bug, the fix, and a draft reply to the reporter - all before I've touched a thing.

4. I say "go" (or don't)

A "GO" reply to the email (or a quick Signal message) triggers the deploy; silence means nothing changes.