skip to content
← work

# memo-flow

indie2025

Run coding agents without babysitting them

tools: Claude Code · memo-garden · Skills CLI · GitHub Issues↗ repo↗ demo
afk-cook overnight loop

## what

Agents need a lot of admin: write the spec, slice the work, ask for tests, watch the context, review. Miss one step and things break later.

memo-flow does that admin for me. It turns an idea into small, well-defined tasks, and the agent finishes them one at a time, tests included. I only step in when something needs my judgment.

One install: npx skills@latest add GuillermoMurillo/memo-flow -a claude-code

20+ skills, 3 hooks, an AFK runner. 10+ projects run on it, including this site.

## why this approach

I like simple. The workflow handles the routine so I can decide what to build and what's wrong with it.

## what breaks

afk-cook runs locally with no isolation. For real blast radius use Sandcastle.

Only GitHub Issues today.

## what i learned

The agent is rarely the problem. A well-defined task ships; a vague one doesn't, no matter how good the model is.

## lessons

  • -hooks enforce what discipline can't.
  • -the agent is rarely the problem. a vague task is.
// read next
project
Memo Garden
Cross-project memory for AI agents. Built on memo-flow.
post
the workflow that works for me
i used to open five agent sessions in parallel and lose half of them
pipeline
building16·live8·apps in the shop24·ai agents & workflows5·web & client builds5·food & hospitality apps4·home automation3·health & learning3·seo & geo2·finance2