Codex + Claude Code automations that ship to your project.
Each card below is a real automation; a focused expert with its own prompt, tools, and model. Works with Claude Code and Codex. Drop the file into your project and your runner picks it up automatically when the task fits.
Specialised agents Claude routes to; automatically.
A subagent is a small Markdown file in .claude/agents/. It carries its own description, tool whitelist, and system prompt. Claude reads the description and decides when to delegate.
Each one runs in its own context window, so your main session stays focused. They can run in parallel, return one structured answer, and never see each other's noise.
- 01
Save the file
Drop the .md into .claude/agents/ at the root of your project, or under ~/.claude/agents/ for personal use across projects.
- 02
Claude reads the description
When you ask Claude something matching, it reads the description field and picks the right subagent.
- 03
Runs in isolation
Its own tools, its own context. Returns one merged answer; no chatter back into your main session.
- 04
You stay focused
Specialised work happens out of your way. Your main session keeps the high-level plan.
Marketing
Outreach, social, content amplification.
Sales & CRM
Pipeline hygiene, prep, outbound, follow-ups.
Content & Docs
Notion, WordPress, Drive, PDFs, sheets.
DevOps & Code
PR review, releases, audits, infra agents.
Research & Intel
Web crawl, market scan, competitor radar.
Production-tested subagents you can drop into a project tonight.
Each card above is a subagent we extracted from a paying-client engagement; prompt, tool list, and operating procedure already pulled apart for re-use. No orchestration code, no DIY guesswork.
You read the description, hit download, drop the .md into .claude/agents/, and Claude routes work to it the next time the description matches a task. Most run end-to-end in under sixty seconds.
Who this is for
- →Engineering leads triaging what to automate next without writing throwaway plumbing.
- →Founders shipping with one engineer plus Claude Code and a tight runway.
- →RevOps and CX teams who want a single subagent rather than ten brittle Zapier zaps.
Three steps to use one
- 01Pick the card whose description matches the team you're trying to help.
- 02Hit download; the .md lands on your machine in one click.
- 03Drop it into .claude/agents/ in your project. Claude routes to it next time the task matches.
Common questions
- What exactly is a Claude Code subagent?
- A single-purpose Claude instance defined by one Markdown file. It carries its own description, tool allowlist, and system prompt. Claude reads the description and decides when to hand work off; you don't orchestrate it manually.
- Do I need to write any code?
- No. The .md file is the whole thing; no library import, no orchestration. Drop it in .claude/agents/ and Claude Code picks it up the next time you ask for that kind of work.
- Will these work with my stack?
- Most subagents talk to HTTP APIs, file I/O, or shell commands; nothing framework-locked. The "stack" line on each card lists the integrations that subagent expects.
- Skill versus subagent; what's the difference?
- A Skill is a slash-command pipeline that orchestrates multiple subagents. A subagent is a single role; either a step inside a Skill or a standalone single-purpose agent. Same Markdown format, different scope.