Process
Six steps, every project
01Agree on what done looks likeBefore anything is built.
02Map the systems and the peopleFind where the data lives, who owns each step, and what can go wrong.
03Build in small stepsEach one checked before the next.
04Prove it worksReal runs and tests, not a status report.
05Keep a person in controlAI and automation do the repetitive work; review and approval stay with people.
06Document the handoffRunbooks, recorded decisions, and a clear next step, so someone else can run it.
AI workspace
Agents as roles, not employees
01One workspaceEvery product, website, and AI role lives in one shared repository, with the rules at the root and one skill library shared by every agent.
02AI rolesEach agent has one job and a written contract: what it owns, what it must not touch, and who it hands work to. I assign the work and make the final call.
03HooksScripts check actions before they run. They block secrets and unrelated files from commits, force pushes to the main branch, and permanent deletes.
04The rulesStay in scope, read the code before making a claim, prove a change works before calling it done, and move files to trash instead of deleting them.
05Memory and handoffsA shared status board and memory files let any role pick up where another stopped. Every handoff includes a brief that can be pasted straight into the next session.
06Public versionA cleaned copy of this setup, with no names or credentials: one prompt that interviews you and builds the same kind of workspace in Claude Code, Codex, or Grok. Agent Workspace Bootstrap on GitHub ↗ (opens in a new tab)