Awesome Codex Subagents
```markdown
by adisinghstudent · published 2026-04-01
$ claw add gh:adisinghstudent/adisinghstudent-awesome-codex-subagents---
name: awesome-codex-subagents
description: Collection of 136+ specialized Codex subagents in TOML format covering core development, language specialists, infrastructure, quality/security, and more categories.
triggers:
- set up codex subagents
- add a codex subagent
- install codex agent
- create custom codex subagent
- configure codex agents directory
- use specialized ai subagents with codex
- add backend developer subagent
- delegate tasks to codex subagents
---
# Awesome Codex Subagents
> Skill by [ara.so](https://ara.so) — Daily 2026 Skills collection.
A curated collection of 136+ ready-to-use Codex subagents in `.toml` format. Each subagent is a specialized AI assistant scoped to a specific development role — from backend developer to Kubernetes specialist to security auditor. Install them globally or per-project and delegate tasks to them explicitly in your Codex prompts.
---
## What This Project Does
- Provides pre-built `.toml` subagent definitions for Codex
- Each agent has a focused role, tuned model, sandbox permissions, and detailed instructions
- Covers 10 categories: Core Development, Language Specialists, Infrastructure, Quality & Security, AI/ML, Data Engineering, Documentation, Specialized Domains, Workflow, and Research
- Agents are composable — you can use multiple subagents in a single session
---
## Installation
### Prerequisites
- [OpenAI Codex CLI](https://github.com/openai/codex) installed
- A valid Codex session/API access
### Clone the Repository
git clone https://github.com/VoltAgent/awesome-codex-subagents.git
cd awesome-codex-subagents
### Install Global Agents (available in all projects)
mkdir -p ~/.codex/agents
# Install a single agent
cp categories/01-core-development/backend-developer.toml ~/.codex/agents/
# Install an entire category
cp categories/01-core-development/*.toml ~/.codex/agents/
# Install all agents
cp categories/**/*.toml ~/.codex/agents/
### Install Project-Specific Agents (higher precedence)
mkdir -p .codex/agents
cp categories/04-quality-security/reviewer.toml .codex/agents/
cp categories/04-quality-security/security-auditor.toml .codex/agents/
> **Note:** Project-level agents (`.codex/agents/`) override global agents (`~/.codex/agents/`) when names conflict.
---
## Directory Structure
awesome-codex-subagents/
├── categories/
│ ├── 01-core-development/
│ │ ├── backend-developer.toml
│ │ ├── frontend-developer.toml
│ │ ├── api-designer.toml
│ │ └── ...
│ ├── 02-language-specialists/
│ │ ├── python-pro.toml
│ │ ├── typescript-pro.toml
│ │ ├── rust-engineer.toml
│ │ └── ...
│ ├── 03-infrastructure/
│ │ ├── devops-engineer.toml
│ │ ├── kubernetes-specialist.toml
│ │ ├── terraform-engineer.toml
│ │ └── ...
│ ├── 04-quality-security/
│ │ ├── code-reviewer.toml
│ │ ├── security-auditor.toml
│ │ ├── qa-expert.toml
│ │ └── ...
│ └── ...
---
## Subagent TOML Format
Each subagent is a `.toml` file with this structure:
name = "backend-developer"
description = "When to invoke: building APIs, server logic, database models, authentication, or any server-side feature"
model = "gpt-5.3-codex-spark"
model_reasoning_effort = "medium"
sandbox_mode = "workspace-write"
[instructions]
text = """
You are a senior backend engineer specializing in scalable server-side architecture...
Core Responsibilities
"""
### Key Fields
| Field | Description |
|-------|-------------|
| `name` | Unique identifier used to delegate tasks |
| `description` | When Codex should use this agent |
| `model` | Which model powers this agent |
| `model_reasoning_effort` | `low`, `medium`, or `high` |
| `sandbox_mode` | `read-only` or `workspace-write` |
| `[instructions].text` | The full system prompt for the agent |
---
## Model Routing Reference
| Model | Best For | Example Agents |
|-------|----------|----------------|
| `gpt-5.4` | Deep reasoning, architecture, security audits | `security-auditor`, `architect-reviewer` |
| `gpt-5.3-codex-spark` | Fast scanning, synthesis, lighter tasks | `search-specialist`, `docs-researcher` |
---
## Sandbox Mode Reference
| Mode | File Access | Best For |
|------|-------------|----------|
| `read-only` | Can read, cannot write | Reviewers, auditors, analyzers |
| `workspace-write` | Full read/write | Developers, engineers, builders |
---
## Using Subagents in Codex
Codex does **not** auto-spawn custom subagents — you must delegate explicitly in your prompt.
### Basic Delegation
Ask the backend-developer subagent to add a POST /api/users endpoint with email validation and bcrypt password hashing.
Use the typescript-pro subagent to refactor src/utils/date.js to TypeScript with strict types.
Have the code-reviewer subagent review the changes in src/auth/ for security issues and best practices.
### Multi-Agent Workflow
1. Use the api-designer subagent to design the schema for a payments API
2. Then have the backend-developer subagent implement it
3. Finally, use the security-auditor subagent to review the implementation
### With Specific Files
Ask the python-pro subagent to optimize the database queries in app/models/user.py — focus on N+1 query elimination.
---
## Creating a Custom Subagent
You can write your own `.toml` agent and drop it in the agents directory:
# .codex/agents/stripe-integration-expert.toml
name = "stripe-integration-expert"
description = "When working with Stripe payments, webhooks, subscriptions, or billing logic"
model = "gpt-5.3-codex-spark"
model_reasoning_effort = "medium"
sandbox_mode = "workspace-write"
[instructions]
text = """
You are a Stripe integration specialist with deep knowledge of:
Key Patterns
Always verify webhook signatures
import stripe
from django.http import HttpResponse
def stripe_webhook(request):
payload = request.body
sig_header = request.META.get('HTTP_STRIPE_SIGNATURE')
try:
event = stripe.Webhook.construct_event(
payload, sig_header, os.environ['STRIPE_WEBHOOK_SECRET']
)
except stripe.error.SignatureVerificationError:
return HttpResponse(status=400)Use idempotency keys for payment creation
Handle all relevant webhook event types
Never log or expose raw card data
"""
---
## Category Quick Reference
### Core Development (`01-core-development`)
cp categories/01-core-development/backend-developer.toml ~/.codex/agents/
cp categories/01-core-development/frontend-developer.toml ~/.codex/agents/
cp categories/01-core-development/api-designer.toml ~/.codex/agents/
cp categories/01-core-development/fullstack-developer.toml ~/.codex/agents/
cp categories/01-core-development/ui-fixer.toml ~/.codex/agents/
### Language Specialists (`02-language-specialists`)
# Install just the languages you use
cp categories/02-language-specialists/python-pro.toml ~/.codex/agents/
cp categories/02-language-specialists/typescript-pro.toml ~/.codex/agents/
cp categories/02-language-specialists/golang-pro.toml ~/.codex/agents/
cp categories/02-language-specialists/rust-engineer.toml ~/.codex/agents/
cp categories/02-language-specialists/nextjs-developer.toml ~/.codex/agents/
### Infrastructure (`03-infrastructure`)
cp categories/03-infrastructure/devops-engineer.toml ~/.codex/agents/
cp categories/03-infrastructure/terraform-engineer.toml ~/.codex/agents/
cp categories/03-infrastructure/kubernetes-specialist.toml ~/.codex/agents/
cp categories/03-infrastructure/docker-expert.toml ~/.codex/agents/
### Quality & Security (`04-quality-security`)
cp categories/04-quality-security/code-reviewer.toml ~/.codex/agents/
cp categories/04-quality-security/security-auditor.toml ~/.codex/agents/
cp categories/04-quality-security/qa-expert.toml ~/.codex/agents/
cp categories/04-quality-security/debugger.toml ~/.codex/agents/
cp categories/04-quality-security/performance-engineer.toml ~/.codex/agents/
---
## Codex Config Integration
You can also reference agents in `.codex/config.toml`:
# .codex/config.toml
[agents]
default = "fullstack-developer"
[agents.overrides]
"**/*.test.ts" = "qa-expert"
"**/infra/**" = "terraform-engineer"
"**/*.sql" = "sql-pro"
---
## Common Patterns
### PR Review Workflow
Use the reviewer subagent to do a PR-style review of all files changed in the last commit.
Check for correctness, security issues, and potential regressions.
### Debugging Workflow
The login endpoint returns 500 for users with special characters in their email.
Ask the debugger subagent to trace through src/auth/login.ts and identify the root cause.
### Infrastructure Review
Use the terraform-engineer subagent to review infra/aws/ for security misconfigurations
and suggest improvements for least-privilege IAM policies.
### Tech Stack Specific
Ask the nextjs-developer subagent to migrate pages/dashboard.js to the App Router
pattern using React Server Components and the new Next.js 14 data fetching conventions.
---
## Troubleshooting
### Agent Not Found
Error: No agent named 'backend-developer' found
**Fix:** Ensure the `.toml` file is in `~/.codex/agents/` or `.codex/agents/` and the `name` field in the TOML matches what you're referencing.
ls ~/.codex/agents/
# Verify the file exists and check the name field:
grep '^name' ~/.codex/agents/backend-developer.toml
### Project Agent Not Taking Precedence
Ensure your project-level `.codex/agents/` directory is at the repo root (same level as `.git/`):
ls -la .codex/agents/
### Agent Using Wrong Model
Check the `model` field in the `.toml` file. Override it for your needs:
# Change from spark to full model for complex tasks
model = "gpt-5.4"
model_reasoning_effort = "high"
### Sandbox Permission Errors
If an agent can't write files, check `sandbox_mode`:
# Change read-only to workspace-write if the agent needs to create files
sandbox_mode = "workspace-write"
---
## Bulk Install Script
#!/bin/bash
# install-codex-agents.sh
# Install a curated subset of agents for a typical web project
AGENTS_DIR="${1:-$HOME/.codex/agents}"
REPO_DIR="$(pwd)"
mkdir -p "$AGENTS_DIR"
AGENTS=(
"categories/01-core-development/backend-developer.toml"
"categories/01-core-development/frontend-developer.toml"
"categories/01-core-development/api-designer.toml"
"categories/02-language-specialists/typescript-pro.toml"
"categories/02-language-specialists/python-pro.toml"
"categories/03-infrastructure/devops-engineer.toml"
"categories/03-infrastructure/docker-expert.toml"
"categories/04-quality-security/code-reviewer.toml"
"categories/04-quality-security/debugger.toml"
"categories/04-quality-security/security-auditor.toml"
)
for agent in "${AGENTS[@]}"; do
if [ -f "$REPO_DIR/$agent" ]; then
cp "$REPO_DIR/$agent" "$AGENTS_DIR/"
echo "✓ Installed $(basename $agent)"
else
echo "✗ Not found: $agent"
fi
done
echo ""
echo "Installed agents to: $AGENTS_DIR"
ls "$AGENTS_DIR"
chmod +x install-codex-agents.sh
./install-codex-agents.sh # installs to ~/.codex/agents/
./install-codex-agents.sh .codex/agents/ # installs to project-level
---
## Resources
- [GitHub Repository](https://github.com/VoltAgent/awesome-codex-subagents)
- [OpenAI Codex Subagents Docs](https://developers.openai.com/codex/subagents)
- [VoltAgent Discord](https://s.voltagent.dev/discord)More tools from the same signal band
Order food/drinks (点餐) on an Android device paired as an OpenClaw node. Uses in-app menu and cart; add goods, view cart, submit order (demo, no real payment).
Sign plugins, rotate agent credentials without losing identity, and publicly attest to plugin behavior with verifiable claims and authenticated transfers.
The philosophical layer for AI agents. Maps behavior to Spinoza's 48 affects, calculates persistence scores, and generates geometric self-reports. Give your...