timkosters

Agent Yolo Build — AI skill for Claude Code

AI community

Run /yolo, walk away, come back to a working MVP.

How to install Agent Yolo Build

This entry records only its repository, not the path inside it, so there is no exact command to give. Open timkosters/agent-yolo-build and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Agent Yolo Build does

Run /yolo, walk away, come back to a working MVP. Autonomous build skill for AI coding agents.

Alternatives in AI

README

/yolo — Autonomous Agent Sessions

Run `/yolo`, walk away, come back to results. Your AI coding agent picks a goal and works on it while you're gone.

Inspired by [the overnight build pattern](https://x.com/bobbyhansenjr/status/2017585853418152313).

Modes

Command What the agent does
/yolo or /yolo build Finds an opportunity and builds a working MVP from scratch
/yolo refactor Finds the highest-leverage improvement in your codebase
/yolo research [topic] Deep research with sources, not just a summary
/yolo explore Reads your notes and projects, surfaces connections you missed

You can also pass a focus: `/yolo build chrome extension` or `/yolo research state of MCP servers`.

Install

Claude Code

mkdir -p ~/.claude/skills/yolo
curl -o ~/.claude/skills/yolo/SKILL.md \
  https://raw.githubusercontent.com/timkosters/agent-yolo-build/main/SKILL.md

Or clone and symlink:

git clone https://github.com/timkosters/agent-yolo-build.git
mkdir -p ~/.claude/skills
ln -s "$(pwd)/agent-yolo-build" ~/.claude/skills/yolo

Other agents

`SKILL.md` is a single markdown file with structured instructions. Copy it into whatever prompt or skill format your agent uses, or paste it as a system prompt. It works with any coding agent that can read files, search the web, and write code.

How it works

  1. Understand (10 min): Scans your projects, notes, and recent work to learn your context
  2. Target (10 min): Researches opportunities (for build) or pain points (for refactor) or sources (for research)
  3. Commit (5 min): Lists 3 candidates, eliminates 2 with reasons, picks one
  4. Work (60-90 min): Does the actual work. Code, tests, iteration.
  5. Present: Summary of what was done, how to use it, what's next

Safety

All modes:

  • No outbound actions (no emails, messages, posts)
  • No paid services or purchases
  • No git push (you decide when to publish)
  • No global installs or syste