How to Install Agents Across 8 AI Platforms
Step-by-step install instructions for agency-agents on Claude Code, Cursor, Copilot, Windsurf, Aider, Continue, Cline, and ChatGPT.
Step-by-step install instructions for agency-agents on Claude Code, Cursor, Copilot, Windsurf, Aider, Continue, Cline, and ChatGPT.
The beauty of the agency-agents library is that it's just text — portable Markdown personas that any modern coding assistant can consume. But every platform has its own quirks for loading system prompts, and the difference between "it works" and "it kind of works" usually comes down to five lines of configuration.
This guide walks through installing agency-agents on eight of the most popular AI coding platforms. Bookmark it.
.claude/agents/ directory (under 30 seconds)Clone the repo once so you have a local copy of every agent:
git clone https://github.com/msitarzewski/agency-agents.git ~/agency-agents
This gives you a folder tree organized by category. Each .md file is a drop-in prompt.
Claude Code has first-class support for agents via the .claude/agents/ directory.
mkdir -p .claude/agents
cp ~/agency-agents/engineering/frontend-developer.md .claude/agents/
Restart your Claude Code session and the agent is available via the agent picker. You can install as many as you like; Claude Code lazy-loads them so there's no performance penalty. See Anthropic's official docs for details.
Cursor uses Rules files located in .cursor/rules/. Copy the agent content into a new rule file:
mkdir -p .cursor/rules
cp ~/agency-agents/engineering/frontend-developer.md .cursor/rules/frontend-developer.mdc
Rename the extension to .mdc and add a type: Always frontmatter block at the top if you want it loaded every session, or type: Agent Requested if you want to invoke it on demand. For the full guide see Install Agency Agents in Cursor.
Copilot reads custom instructions from .github/copilot-instructions.md. The catch: Copilot only supports one instruction file per repo, so you'll need to either pick your favorite agent or concatenate several.
mkdir -p .github
cat ~/agency-agents/engineering/frontend-developer.md > .github/copilot-instructions.md
For multi-agent workflows on Copilot, see the full walkthrough in Install Agency Agents in Copilot.
Windsurf uses Cascade memories. Open Cascade settings, create a new memory with the agent content, and mark it as "Always active." Windsurf's memory system supports multiple concurrent personas, so you can install several agents and switch between them via chat commands.
# Windsurf has no file-based install, use the UI
open -a Windsurf
Aider supports conventions via a .aider.conf.yml file and can accept system prompts via the --message or --read flags. The cleanest approach is to drop the agent content into a file and reference it:
# .aider.conf.yml
read:
- ~/agency-agents/engineering/frontend-developer.md
Restart Aider and the agent is part of every session's context.
Continue uses a config.json file in ~/.continue/. Add a custom assistant that references the agent file:
{
"assistants": [
{
"name": "Frontend Developer",
"systemMessage": "file:~/agency-agents/engineering/frontend-developer.md"
}
]
}
Continue will hot-reload the config and the agent appears in the assistant picker.
Cline (formerly Claude Dev) supports custom instructions through its settings panel. Paste the agent content into the "Custom Instructions" field. Cline persists this across sessions and will apply it to every task.
For multi-agent workflows on Cline, consider using the Agents Orchestrator pattern, where a master agent delegates to sub-agents by swapping instructions mid-session.
ChatGPT is the odd one out — it doesn't support repository-level config, but you can create a Custom GPT per agent. Go to the GPT builder, paste the agent content into the instructions field, and save. You'll need to create one GPT per agent, which is tedious but gets the job done.
The downside of ChatGPT is that you lose the ability to compose agents within a single session. Multi-agent orchestration works much better on platforms with native config files.
For sheer flexibility, Claude Code wins. The .claude/agents/ directory was designed with multi-persona workflows in mind, and loading 20+ agents takes seconds. Cursor is a close second if you're already in the JetBrains/VS Code ecosystem. Copilot trails because of the single-file limitation.
If you're just getting started, pick the platform you already use daily. Agent quality matters more than platform choice. Read How Agents Beat Prompt Engineering in 2026 for the "why" behind this.
After installing, run this quick test: ask the agent a question that only a specialist would answer well. For the Frontend Developer, try "Should I use <button> or <div onclick> for this toggle?" A correctly installed agent will explain the accessibility difference and recommend <button> immediately. If you get a mealy-mouthed "it depends" response, the install probably didn't load.
On Claude Code, Cursor, Continue, and Windsurf, yes. On Copilot and ChatGPT, you're limited to one at a time per project.
Yes. The simplest workflow is git pull in your local clone, then re-copy the updated files. Automate it with a shell alias if you do this often.
No. Agents only load into context when you invoke them. Idle agents cost zero tokens.
Yes. They're just Markdown. Edit them freely. If your edits are useful to others, send a PR upstream.
Claude and ChatGPT mobile apps accept custom instructions via their settings. The workflow is the same as ChatGPT Custom GPTs above.
Eight platforms, one library, fifteen minutes of setup total. Once you've done it, you'll have an army of 150 specialists available in every coding session. The only question left is which agent to invoke first.
Browse all 150 agents at aiskill.market/agents or submit your own skill.
AI soul and personality tracing skill — gives your AI agent a consistent persona, tone, and conversational identity across all interactions. 319.2K installs.
Official Microsoft bundle: Azure AI, deploy, hosted Copilot SDK, observability, and cloud migration. 600K+ combined installs.
Design short-term, long-term, and graph-based memory architectures for agents
Official Agent Skills suite from ElevenLabs: text-to-speech, speech-to-text, voice agents, sound effects, music, and voice isolation. Works in Claude Code, Cursor, and other compatible agents.