The entire dev org chart, compressed into TOML

What it does

This repository is a curated library of over 160 Codex subagents—specialized AI assistants packaged as .toml configuration files. Each file bundles a role, domain-specific instructions, a pre-selected model tier, and a filesystem sandbox mode. They are designed to live in Codex’s agent directories, letting you invoke a specialist rather than hoping the generalist guesses right.

The interesting bit

The maintainers did the tedious tuning work: mapping each role to a specific model (e.g., gpt-5.4 for security audits, gpt-5.3-codex-spark for quick scans) and locking down sandbox permissions so reviewers stay read-only. It effectively turns Codex from a solo developer into an entire department with job descriptions.

Key highlights

• 166+ subagents across 13 categories, from core development to infrastructure and quality/security
• Each agent ships with a tuned model field and a sandbox_mode (read-only for auditors, workspace-write for builders)
• Narrow specializations like terragrunt-expert, powershell-5.1-expert, and ui-fixer are included
• Aligned with OpenAI’s official Codex subagent docs and directory precedence rules
• Project-specific agents automatically override global ones when names collide

Caveats

• Codex will not auto-spawn these; you must explicitly delegate to them in prompts
• The repo is strictly a configuration library—no runtime, orchestration, or automation beyond the TOML files

Verdict

Grab it if you use Codex and want pre-tuned experts for your stack. Ignore it if you are looking for a framework or agent runtime; this is a parts catalog, not an engine.