Building, guiding, and cleaning up parallel Codex agents on Omarchy.

Set this up once and you can spin up a fresh worktree, brief Codex, launch isolated dev services, and unwind everything in under a minute.

I’m Thomas Rice, Co-Founder of Minotaur Capital. Our primary codebase is called “Taurient”, which is why the examples use taurient in paths and filenames. Swap those references for your own repo and directory structure as you copy the snippets across.

Follow the docs in order:

1. Shell + Worktree Helpers – add aliases, wt/dwt, and the optional newterm launcher.
2. Dev Server Automation – create ~/dev.sh, add the baseline /etc/hosts entry, and wire nginx to the generated configs.
3. Agent Briefings – keep reusable AGENTS.md, PROJECT.md, and PROJECT_AI_NOTES.md templates in every worktree.
4. (Optional) UI Extras – Hyprland workspace labels, Waybar modules, and a voice-toggle binding.

Each document tells you exactly which file to edit, what to paste, and which bits are optional.

1. Spawn a workspace: wt <name> clones a Git worktree, seeds Poetry, and (optionally) renames the Hyprland workspace so Waybar shows the new label.
2. Brief the agent: Update PROJECT.md, keep long-lived guidance in AGENTS.md, then run cproj to launch Codex in high-reasoning mode.
3. Launch infra: tdev allocates a deterministic port, tags /etc/hosts, writes an nginx vhost, and opens a terminal for the Taurient app server.
4. Develop: Iterate on the brief while Codex and your dev servers do the work; toggle voice input with SUPER+S if you keep the optional binding.
5. Clean up: dwt <name> prunes the worktree, removes nginx + hosts artefacts, clears saved port metadata, and resets the workspace label.

• Deterministic hostname + port per worktree (<branch>.t.com, hashed port offsets) so parallel branches never collide.
• A repeatable agent briefing workflow—Codex always reads your stack rules and the current plan before it writes code.
• Optional desktop polish (Waybar, Hyprland, voice toggle) that makes multi-agent work visible at a glance.

• Replace the placeholder paths (e.g. $HOME/minotaur/taurient) with your own repo layout before running the scripts.
• Swap the terminal launcher (newterm) or the app entry points in dev.sh to match your toolchain.
• Skip any optional sections if you do not use Hyprland/Waybar or voice input—the core workflow only needs the shell helpers, dev.sh, and the briefing files.
• Add PROJECT.md and PROJECT_AI_NOTES.md to your .gitignore so the briefs and scratchpad never leave your workstation.

Once everything is wired up, the flow is:

wt my-new-feature
# edit PROJECT.md (and optionally PROJECT_AI_NOTES.md)
cproj

Questions or tweaks you’d like to see? Fork it, adapt it, and keep refining your multi-agent workflow.