Eden agent skill for Codex, Claude Code, and Cursor

Who this is for

This is a power-user add-on, not part of normal setup. Most people connect Claude, Cursor, or ChatGPT and chat — they don't need it. You want the skill when an autonomous coding agent is calling Eden's tools on its own and you want it to behave: stay connected, schedule to the right platforms, and handle failures gracefully.

If you just want to connect an assistant, start with the quick start instead.

What it fixes

Coding agents tend to stumble on the same three things. The skill addresses each:

• Connection drops. When the OAuth session expires or the server restarts, an agent often sees a cryptic OAuth authorization required and either spins or invents a workaround. The skill tells it the truth: reconnect the MCP and retry — it's a connection issue, not a bad request.
• Scheduling mistakes. Per-platform rules (Instagram/TikTok need media, YouTube is Shorts-only with a title, LinkedIn PDFs, X/Threads threads) and editing a post in place instead of leaving a broken draft behind and cloning a "clean" copy.
• Error handling. Eden returns a structured { ok: false, status, message } for every expected failure. The skill maps each status to the right move — fix the input, retry once, or surface it — so the agent stops escalating around clean errors.

Download

Install it

The skill is a folder named eden-mcp/. Unzip it into your agent's skills directory:

1. 01

   Claude Code

   Unzip into ~/.claude/skills/ (available everywhere) or a project's .claude/skills/. Claude Code picks it up automatically — the skill loads when a task involves the Eden MCP.

2. 02

   Codex

   Unzip into your Codex skills/prompts directory (e.g. ~/.codex/skills/). Point the agent at eden-mcp/SKILL.md so it reads the router first, then the reference file that matches the task.

3. 03

   Cursor

   Drop the eden-mcp/ folder into your project (for example under .cursor/skills/) and reference SKILL.md from your rules so the agent loads it alongside the Eden MCP server.

The skill assumes the Eden MCP is already connected. If it isn't, connect it first — connect Claude, Cursor, or ChatGPT.

A note on staying connected

The most common failure an agent hits is an expired connection mid-run, which surfaces as OAuth authorization required or Unauthorized. That's almost always a reconnect, not a code or argument problem — re-authorize Eden in your client and retry. The skill makes the agent do exactly that instead of looping. See also refreshing tools.