Two kinds of markdown file sit behind AgentMore, and it is worth knowing which is which.

​
The setup file — start here

Fetch and execute the appropriate instructions to set me up for AgentMore from https://agentmore.app/agent-setup/prompt.md
/agent-setup/prompt.md is a router. It carries one section per kind of agent — the Claude Code plugin, the Codex marketplace manifest, the CLI for anything with a shell, the MCP server for anything without one — and the agent reading it picks its own. That is the only address most people need. It is plain markdown; open it and read it yourself before you run it.

​
Skill files — what a job looks like

A skill file is a readable SKILL.md describing how one job is done: what to call, in what order, and when a call spends money. Your agent loads one when the task calls for it.
Skill fileUse it whenURL
The collectionYou want the agent to see every skill and pickhttps://agentmore.app/skill.md
Short aliasYou are typing the collection address by handhttps://agentmore.app/SKILL.md
UI skillsYou are working on interface design and polishhttps://agentmore.app/ui/SKILL.md
Gauntlet LoopYou want the loop run against external targets onlyhttps://agentmore.app/gauntlet-loop/SKILL.md
Put set up in front of any of those and send it.
Loading the collection directly is not a substitute for the setup line at the top of this page. It teaches your agent what the skills are without installing the CLI or asking you for a key, so the agent ends up knowing what it could do and unable to do it. Use the setup line first; reach for an individual skill file to narrow an already-configured agent to one job.
The collection is also served at https://agentmore.app/agentmore/SKILL.md, and on the old noctrn.xyz host at both /skill.md and /noctrn/SKILL.md. Every one returns the same bytes, so a setup line saved before the rename keeps working. Use the short address above for anything new.

​
What your agent does with the setup file

1

It fetches the file

An ordinary HTTP GET. Reading costs nothing and needs no account.
2

It finds its own section

Claude Code installs a plugin; an agent with a shell installs the CLI; an agent without one is pointed at the MCP server. You do not have to know which you have.
3

It asks for a key

Running an endpoint needs an API key. Create one at API keys and paste it when asked.
4

It works

Send a real task and watch it go:
Find the last 50 posts for this hashtag on TikTok and tell me what is working.
Discovery and inspection are free and need no key — an agent can search the catalog and read an endpoint’s schema and price before you have signed up. Only running spends.

​
Next

Set up your agent

Per-agent steps for Claude Code, Codex, Cursor and the rest.

API keys

Create the key that lets a call actually run.

MCP server

For agents that connect over Model Context Protocol instead.

Spend caps

Set a per-call and per-day ceiling before you run anything.