dotcodex

My Codex setup, just the coding parts.

Contents

• instructions/SYSTEM.custom.md is the system prompt I use to customize model behavior. The default prompt casts Codex as an eager-to-please old friend. Most of my changes make that friend think before nodding along and keep it from inventing a framework whenever a screwdriver would do.
• instructions/SYSTEM.default.md is included as a baseline for comparison. View the system prompt changes.
• instructions/AGENTS.md is for more personalized rules and gotchas. The "User Memory" it references is not included because it contains personal context, and it's unrelated to coding anyway. I curate memories manually so I can decide exactly what is retained and loaded into context.
• skills/ contains only the skills I use for coding and related work.
• config.toml has the bits of my Codex config that seemed worth sharing.

Automations are too specific to my workflows and environment, so they're not included here.

Related projects

These projects are also part of this setup but live in dedicated repositories:

• ⛳️ Questline provides a workflow for developing software one bounded, reviewable deliverable at a time.
• 🪏 Knotbane finds and simplifies cyclomatic complexity hotspots in PHP code.
• I also have agent-ready Laravel starters for applications and packages.

Automatic skill invocation

My skills are designed to be selected automatically from the request, not treated as a command palette. Explicitly invoking a skill remains available, but it is an override rather than the usual interface.

That differs from some popular collections such as Matt Pocock's skills for Claude, which favor user-invoked workflows. Every skill here is triggered by the model instead. Even workflows like Questline are typically driven using special vocabulary like "waypoint" and "compass" rather than slash commands.

View the system prompt changes

git diff --no-index -- instructions/SYSTEM.default.md instructions/SYSTEM.custom.md