Skip to content

Introduction

Where the guides start: the principles, config, workflow and skills a developer of ten years put around coding agents.

These docs describe the rules, hooks and skills I put around coding agents while handing them work, and why each has the shape it has. They do not cover first steps or installation. Each guide explains how a piece works, then shows my config.

Three harnesses share the rules, hooks and skills of one repoOne repo, agent-config, holds rules, hooks and skills. Rules are text the agent reads and follows, hooks stop a command that breaks a rule before it runs, and skills set the order of the work. Claude Code, opencode and pi all use the three.agent-configone repoRulesfollowedHooksblockSkillsset stepsClaude Code · opencode · piall three share them

What is covered

  • Principles — the principles under this whole setup and the incidents behind them
  • Agent config — rules, hooks, and subagents for Claude Code, pi, and opencode, and the plugins and MCP servers I keep on
  • Workflow — how a task goes from plan to merged PR
  • Skills and habits — the skills I call beside each workflow step, and the habits that stuck
  • Knowledge loop — how a finished task's record is distilled into the knowledge base and comes back to the next plan
  • Herdr — the terminal workspace that holds every agent session

Principles

Principles under my agent setup: enforce rules with structure, give a checkable criterion first, let the model fill in the verdict, fix mistakes at the source.