This is my current Codex setup.
It contains the rules, docs, and reusable skills that shape how the agent behaves when we work together: how tasks are approached, which tools to prefer, and how changes are checked before handoff.
-
AGENTS.md
Core behavior contract for the agent. -
TOOLS.md
Practical guidance for the command-line tools used in this workflow. -
skills/
Task-specific skill packs. Each skill has its ownSKILL.mdwith focused instructions. -
docs/
Supporting runbooks and references used during execution. -
SYSTEM_PROMPT.md
Canonical Codex system override source file. -
scripts/
Home for executable helpers used by this toolkit.
This is where I keep the common scripts that agents can self-improve over time. Still WIP.
scripts/docs.ts
Scans docs folders, reads front matter, and prints a docs index withsummaryandread_whenmetadata.
This is what powers thengents docscommand.
The point of this directory is to keep small, practical automation close to the instructions and docs it supports.
- People building their own Codex/agent workflow.
- Anyone who wants a modular setup with docs + skills + tool conventions in one place.
This is largerly inspired by @steipete's agent-scripts.