claude-agents

Claude Agents — Multi-Agent Marketplace for Claude Code

A unified marketplace for Claude Code that brings together multiple AI coding agents. Currently includes:

• Cursor Agent — delegate coding tasks, code reviews, and investigations to Cursor Agent
• Codex — OpenAI's coding agent for code review and task delegation

Prerequisites

• Claude Code CLI installed
• For Cursor Agent: cursor-agent CLI installed and authenticated (cursor-agent login)
• For Codex: codex CLI installed and authenticated (codex login)
• Node.js 18.18 or later

Installation

In Claude Code, run the following commands:

/plugin marketplace add yangtau/claude-agents-plugins
/plugin install cursor@claude-agents
/plugin install codex@claude-agents
/reload-plugins

Verify with:

/cursor:setup
/codex:setup

---

Cursor Agent Commands

Command	Description
/cursor:setup	Check cursor-agent availability and auth status
/cursor:rescue <prompt>	Delegate investigation, fixes, or follow-up rescue work to Cursor Agent
/cursor:review [focus]	Run a code review using Cursor Agent
/cursor:model	List models, or set the workspace default model (/cursor:model <id>; --clear to reset)
/cursor:status [job-id]	Show active and recent jobs
/cursor:result [job-id]	Show output of a completed job
/cursor:cancel [job-id]	Cancel a running background job
/cursor:execute-plan [path]	Execute a Claude Code plan using Cursor Agent

Cursor Rescue Examples

# Basic rescue request
/cursor:rescue "refactor the auth module to use JWT"

# With a specific model (overrides workspace default for this run only)
/cursor:rescue --model grok-4-20 "add error handling to the API layer"

# Set default model for this workspace (then omit --model on rescue runs)
/cursor:model grok-4-20

# Continue the previous rescue thread
/cursor:rescue --resume "continue the previous task"

# Run in background
/cursor:rescue --background "migrate the database schema"

# Execute a Claude Code plan (auto-detects latest plan in ~/.claude/plans/)
/cursor:execute-plan
# Or specify a plan file explicitly
/cursor:execute-plan ~/.claude/plans/my-plan.md

---

Codex Commands

Command	Description
/codex:setup	Check Codex availability and auth status
/codex:rescue <prompt>	Delegate investigation, fixes, or follow-up rescue work to the Codex rescue subagent
/codex:review	Run a code review using Codex
/codex:adversarial-review [focus]	Run an adversarial code review using Codex
/codex:model	List models, or set the workspace default model (/codex:model <id>; --clear to reset)
/codex:status [job-id]	Show active and recent jobs
/codex:result [job-id]	Show output of a completed job
/codex:cancel [job-id]	Cancel a running background job
/codex:execute-plan [path]	Execute a Claude Code plan using Codex

Codex Rescue Examples

# Basic rescue request
/codex:rescue "refactor the auth module to use JWT"

# With a specific model (overrides workspace default for this run only)
/codex:rescue --model gpt-5.3-codex-spark "optimize the database queries"

# Set default model for this workspace (then omit --model on rescue runs)
/codex:model spark

# With reasoning effort level
/codex:rescue --effort high "design a new caching layer"

# Run in background
/codex:rescue --background "migrate the database schema"

# Resume last conversation
/codex:rescue --resume "continue the previous task"

# Run code review
/codex:review

# Execute a Claude Code plan (auto-detects latest plan in ~/.claude/plans/)
/codex:execute-plan
# Or specify a plan file explicitly
/codex:execute-plan ~/.claude/plans/my-plan.md

---

Architecture

.claude-plugin/marketplace.json     Marketplace registry
plugins/
  cursor/                           Cursor Agent plugin
    .claude-plugin/plugin.json      Plugin manifest
    commands/                       8 slash commands
    agents/cursor-rescue.md         Auto-delegation subagent
    skills/                         Internal skills
    scripts/                        Companion scripts
  codex/                            Codex plugin (OpenAI)
    .claude-plugin/plugin.json      Plugin manifest
    commands/                       9 slash commands (including execute-plan)
    agents/codex-rescue.md          Auto-delegation subagent
    skills/                         GPT-5.4 prompting, result handling
    scripts/                        Codex companion scripts

How It Works

All commands invoke the respective CLI tools in headless mode under the hood:

• Cursor: cursor-agent -p --trust
• Codex: codex --approval-mode full-auto (or similar)