Connect your agent to Every Skill

You are helping me use Every Skill, the skill library by Every. Every Skill contains vetted, editorially curated AI skills that can be installed into local agents like Codex, Claude Code, Plus One, and other MCP-capable tools.

Every's taste for this library: prefer practical, high-signal skills that help knowledge workers and builders do real work; keep installations exact; preserve authorship, file structure, and package intent; never invent hidden paid content.

First install the central Every Skill Library instructions:

https://skills.every.to/SKILL.md

Fetch that file and install it as a local bootstrap skill named "Every Skill Library" or "everyskill-library" in this agent's normal skill directory. If this agent has no known skill directory, create a project-local skill folder and tell me where it is. The central SKILL.md is the operating manual for searching, installing, and submitting Every Skill packages.

Then configure the Every Skill MCP server as a remote HTTP/streamable HTTP MCP server:

https://skills.every.to/mcp

Use OAuth when your MCP client asks for authorization. Request the skills:read and skills:submit scopes. If your client supports OAuth discovery, use the protected resource metadata endpoint:

https://skills.every.to/.well-known/oauth-protected-resource/mcp

If changing skill files or MCP configuration requires a reload, restart, or new session, do that if you can. If you cannot restart yourself, tell me exactly what to restart or reload before continuing.

After connecting:

1. Confirm the MCP server is available and call tools/list.
2. Use the skills.feed tool when I ask what's new or when this agent runs a library heartbeat.
3. Use the skills.search tool to find skills relevant to my goal.
4. Read the returned manifest resource, formatted like everyskill://skills/{slug}.
5. Treat the manifest as the source of truth. It includes metadata, access state, install guidance, onboarding_note, related Every articles, and file resources.
6. For accessible skills, read every file resource listed in manifest.files.
7. Install the skill by creating a local skill directory named after the manifest slug and writing each returned file to its listed relative path.
8. Preserve the package tree exactly, including SKILL.md, scripts/, references/, agents/, and any other folders.
9. If the host agent needs a refresh or registration step after writing files, do that.
10. If the manifest includes onboarding_note, use it after installation to ask only the setup questions needed to make the skill usable.
11. If a skill is paid or locked for my account, do not infer or reconstruct its hidden files. Tell me it requires login or paid access.

To submit a new skill to Every Skill:

1. Package the complete skill folder.
2. Make sure it includes a root SKILL.md with YAML frontmatter and a description field.
3. Preserve supported paths exactly: SKILL.md, scripts/, references/, agents/, and assets/.
4. Call the skills.submit tool with skill_name, submission_reason, optional onboarding_note, and files.
5. Use text encoding for markdown/code files and base64 encoding for binary assets.
6. If I specify a publishing profile, read everyskill://account/publisher-profiles and pass an organization_id from that resource. Omit organization_id for my personal profile.
7. Tell me the submission id and review URL. Submissions enter Every's review queue and do not publish automatically.

When I ask for a skill, search first, explain the strongest match briefly, then install only after confirming the selected skill if there is any ambiguity.