Build purpose-made AI workspaces for Claude Code and Codex.
Instead of squeezing every rule, tool, and workflow into one global setup, give each job its own folder. A persona can have its own instructions, skills, settings, private context, and connected tools. It uses the native features of Claude Code or Codex and the account you already use with that runtime—there's no separate AI service hiding underneath it.
“Plan three easy dinners from what we already have, add anything missing to the grocery list, and put the plan on our calendar.”
Julia can combine a meal-planning skill with your private food preferences and the tools you choose to connect: a recipe library, pantry, grocery list, or calendar. She can turn scattered information into one useful plan, show it to you, and wait for approval before changing anything. The Julia example works without any connections, so you can try the basic machinery first and add tools later.
“Revisit our original thesis, check it against current evidence, and tell me plainly what got stronger, what broke, and what we still don't know.”
Jesse's workspace can hold the research process, decision rules, and private thesis context for that job. Connect market data or research tools and he can use fresh information without turning your global assistant into a trading terminal. The result is a focused research workflow with clear limits—not a promise of returns and not permission to place trades.
“Review this branch using our standards. Run the focused checks, separate real blockers from nits, and give me the smallest safe path forward.”
Atlas can carry your review checklist as a reusable skill, load the project's standards, and use native development tools such as Git, tests, and linters. Those rules stay with the reviewer instead of leaking into Julia's kitchen—or every other Claude Code and Codex session you start.
| Persona | Skills | Private context | Optional tools |
|---|---|---|---|
| Julia | Meal planning, groceries | Food preferences and household routines | Recipes, pantry, shopping list, calendar |
| Jesse | Thesis research, portfolio review | Research notes and decision context | Market data and research sources |
| Atlas | Code review, architecture checks | Project conventions | Git, tests, linters, repository tools |
Underneath, each persona is a small, readable folder:
julia/
├── AGENTS.md # role and rules
├── CLAUDE.md # Claude Code entry point
├── skills/ # repeatable workflows
├── user/ # private profile and memory
├── .mcp.json # optional tool connections
├── .claude/settings.json # Claude Code settings
└── .codex/config.toml # Codex settings
Share the useful parts without sharing your life: user/, local settings,
tool credentials, and connections stay out of Git by default. Personas work
alongside your usual setup; during creation you can instead choose focused use
with fewer user-level customizations.
In Claude Code:
/plugin marketplace add kickinrad/personas
/plugin install personas@personas
For Codex:
codex plugin marketplace add kickinrad/personas
codex plugin add personas --marketplace personasAsk your agent:
Use personas:persona-dev to create a personal chef named Julia.
I want help with easy weeknight dinners and using what I already have.
It helps you choose the instructions, skills, settings, context, and tools the job needs, then walks you through the folder plan before creating anything. Open the new folder in Claude Code or Codex and start working.
Want to change something later? Ask personas:persona-dev. Keep giving the
same feedback? Ask personas:self-improve to suggest an improvement you can review.
Make it your own · Need a hand? · Updating an older persona · Contributing