Maintainer of this skill?
by bodencrouch
Route Codex and Claude Code subagents to the right model and reasoning level, with ready-made worker definitions
# Add to your Claude Code skills
git clone https://github.com/bodencrouch/codex-better-subagentsGuides for using ai agents skills like codex-better-subagents.
See how codex-better-subagents compares with popular alternatives.
codex-better-subagents is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by bodencrouch. Route Codex and Claude Code subagents to the right model and reasoning level, with ready-made worker definitions. It has 0 GitHub stars.
codex-better-subagents's catalog security scan is still queued. You can run an instant dependency and prompt-injection check now with the "Scan for vulnerabilities" button above.
Clone the repository with "git clone https://github.com/bodencrouch/codex-better-subagents" and add it to your Claude Code skills directory (see the Installation section above).
codex-better-subagents is primarily written in Shell. It is open-source under bodencrouch on GitHub, so you can review or fork the full source.
Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh codex-better-subagents against similar tools.
No comments yet. Be the first to share your thoughts!
Available after the catalog security scan passes.
⚠️ Third-Party Software Notice
This skill is third-party open-source software developed and hosted independently on GitHub. SkillsLLM is an informational directory and does not control or maintain the underlying repository.
Any security checks, ratings, or warnings displayed by SkillsLLM are automated and limited in scope. They do not constitute a security certification or guarantee that the software is safe, error-free, or free from malicious code, vulnerabilities, compromised dependencies, or prompt-injection risks.
Review the source code, permissions, dependencies, and configuration before installing or running any third-party skill. Use is at your own risk. To the maximum extent permitted by applicable law, SkillsLLM is not liable for losses arising from third-party software.
The deep catalog scan for this skill is still queued. Run an instant dependency check now instead.
A routing policy, an installable skill and four worker definitions for choosing a model and reasoning level for each delegated job.
The guide also records native-handoff failures observed in the author's cross-vendor proxy setup and a CLI fallback. Those observations are harness-specific, not claims about every Codex or Claude Code installation.
| Path | What it is |
|---|---|
skills/codex-better-subagents/SKILL.md |
Discoverable routing skill with tier boundaries, explicit settings and a dispatch example |
ROUTING.md |
The guide: four tiers, how to set reasoning, native spawns, CLI workers and task packets |
agents/claude/ |
Claude Code subagents: basic-worker, standard-worker, deep-worker, frontier-worker |
agents/codex/ |
The same four workers as Codex agent roles |
codex/worker.config.toml |
A lean Codex profile for headless codex exec -p worker runs |
.claude-plugin/ |
Claude Code plugin and marketplace manifests |
install.sh |
Copies everything into place and keeps a timestamped backup of any file it replaces |
This repo is also a plugin marketplace. In Claude Code:
/plugin marketplace add bodencrouch/codex-better-subagents
/plugin install codex-better-subagents@codex-better-subagents
On Claude Code 2.1.275 or later, one step works too:
/plugin install codex-better-subagents --marketplace bodencrouch/codex-better-subagents
The plugin adds the routing skill and four Claude workers as codex-better-subagents:basic-worker and so on. It does not install ROUTING.md or the Codex files; use install.sh for those.
git clone https://github.com/bodencrouch/codex-better-subagents.git
cd codex-better-subagents
./install.sh
This writes:
~/.agents/rules/subagent-model-routing.md~/.agents/skills/codex-better-subagents/SKILL.md~/.claude/agents/{basic,standard,deep,frontier}-worker.md~/.codex/agents/{basic,standard,deep,frontier}-worker.toml~/.codex/worker.config.tomlTo stage an installation elsewhere without changing your home configuration:
stage_root="$(mktemp -d)"
CODEX_BETTER_SUBAGENTS_HOME="$stage_root" ./install.sh
The installer backs up changed files with unique timestamped filenames. Review those backups when updating an existing setup.
Then add one line to your global AGENTS.md or CLAUDE.md so agents read the guide before they delegate:
Before you delegate, read `~/.agents/rules/subagent-model-routing.md`.
In Claude Code, pick a worker by type:
Agent(subagent_type: "basic-worker", prompt: "List every file that imports parse_config and report file:line.")
In a Codex harness exposing task_name and fork_turns, pass a configured role and fresh context:
spawn_agent(task_name: "bounded-fix", agent_type: "standard-worker", fork_turns: "none", message: "<self-contained task packet>")
Match the call to your tool schema: some harnesses expose fork_context: false instead. Installing role files does not guarantee that every harness registers custom roles automatically; check its available role list.
From a terminal, when native slots run out or a spawn fails:
claude -p --model claude-haiku-5-5 --effort high --no-session-persistence --output-format json < task.md
codex exec -p worker -m gpt-6-luna -c model_reasoning_effort=high --ephemeral - < task.md
See ROUTING.md for the full rules, including parallel CLI workers.
The model IDs and price examples reflect the author's October 2026 configuration and may be proxy aliases. They are not guarantees of provider availability or current pricing. If your provider offers different models, edit the tier table in ROUTING.md and the model lines in agents/. Keep the tier names, because the agent files and the guide refer to them.
The codex/worker.config.toml profile turns off plugins, apps, memories and hooks. To keep workers fast, add an enabled = false block to ~/.codex/worker.config.toml for each MCP server in your ~/.codex/config.toml. That keeps the servers on in your normal sessions.
The standalone worker.config.toml loading convention is supported by the author's installed CLI; other Codex versions may use [profiles.worker] in the main configuration instead. Check codex exec --help before using -p worker, and enable only feature flags your version supports.
The skill was evaluated before publication with five no-guidance controls and five fresh-context guided runs. Browser-tier selection matched this policy in 1/5 controls and 5/5 guided runs. Three additional guided model smoke checks passed. This measures dispatch-planning compliance with this policy, not model quality, execution success or cost savings. See the validation record.
Issues and pull requests are welcome. Measured results help most: which spawns worked or failed, on which harness version, with which models. See CONTRIBUTING.md.