# Aeon

> Aeon is an autonomous agent framework. It runs your own skills on a schedule in GitHub Actions, keeps state in git-committed memory, and dispatches every run through one harness adapter that works across nine coding-agent CLIs (Claude Code, Codex, Grok, Kimi, Pi, Vibe, fx, Cursor, Hermes). GLM Coding Plan is a Claude AI Gateway provider, not a harness. Skills are Markdown with a spec-form frontmatter; a least-privilege `requires:` -> dashboard vault contract wires secrets. Start it from the browser with Aeon Connect (https://www.aeon.fun/connect: sign in with GitHub, connect a model, pick skills), or clone the repo and run `./aeon init`; either way it runs itself on your own GitHub Actions. Once it exists, drive it from ChatGPT, Claude or Cursor through the hosted MCP server at https://www.aeon.fun/connect/mcp.

Repo: https://github.com/aeonfun/aeon
Site: https://www.aeon.fun

## Core entry points

- [AGENTS.md](https://raw.githubusercontent.com/aeonfun/aeon/main/AGENTS.md): Start here. The instance operating manual an Aeon agent reads at runtime - how Aeon works, strategy, voice/soul hierarchy, memory, tools, capability mode, skill chaining, notifications, network/secrets, security.
- [CLAUDE.md](https://raw.githubusercontent.com/aeonfun/aeon/main/CLAUDE.md): The same operating manual, Claude Code loads it automatically.
- [STRATEGY.md](https://raw.githubusercontent.com/aeonfun/aeon/main/STRATEGY.md): The instance north-star: priorities, audience, hard constraints, what to optimize for and avoid.
- [aeon.yml](https://raw.githubusercontent.com/aeonfun/aeon/main/aeon.yml): The operator config: which skills are enabled, their schedules, vars, per-skill model/harness overrides, chains, and reactive triggers. The workflow that runs them is `.github/workflows/aeon.yml`.

## Setup and configuration

- [Aeon Connect](https://www.aeon.fun/connect): The fastest start, in the browser. Sign in with GitHub, create your instance repo (a public fork or a private copy of aeonfun/aeon), install the Aeon Connect GitHub App on it, connect a model, pick skills. No clone or terminal.
- [Hosted MCP server](https://www.aeon.fun/connect/mcp): Drive your agent from ChatGPT, Claude, Claude Code or Cursor. Add the URL as a custom connector (Claude Code: `claude mcp add --transport http aeon https://www.aeon.fun/connect/mcp`), sign in with GitHub and pick the agent repo. List and run skills (on the repo's own GitHub Actions), read runs, output and memory, edit skills, strategy and soul. Secret values are never readable or writable over it.
- [docs/aeon-setup.md](https://raw.githubusercontent.com/aeonfun/aeon/main/docs/aeon-setup.md): The setup skill - get an instance running from scratch from a chat in your coding agent.
- [docs/CONFIGURATION.md](https://raw.githubusercontent.com/aeonfun/aeon/main/docs/CONFIGURATION.md): Configuration and advanced reference.
- [docs/CORE.md](https://raw.githubusercontent.com/aeonfun/aeon/main/docs/CORE.md): The Core - the ~15 load-bearing skills (self-evolution, self-healing, fleet) that make Aeon autonomous rather than just scheduled.
- [docs/CAPABILITIES.md](https://raw.githubusercontent.com/aeonfun/aeon/main/docs/CAPABILITIES.md): Skill capabilities taxonomy - read-only vs write vs capability mode.

## Skills and packs

- [catalog/skills.json](https://raw.githubusercontent.com/aeonfun/aeon/main/catalog/skills.json): Machine-readable catalog of every bundled skill - slug, description, category, requires, mcp, install command.
- [docs/skill-packs.md](https://raw.githubusercontent.com/aeonfun/aeon/main/docs/skill-packs.md): Skill packs - opt-in bundles so a fork only runs the handful of skills it wants.
- [docs/community-skill-packs.md](https://raw.githubusercontent.com/aeonfun/aeon/main/docs/community-skill-packs.md): Publish and install community skill packs.
- [docs/skill-integrity.md](https://raw.githubusercontent.com/aeonfun/aeon/main/docs/skill-integrity.md): The eyebrow lockfile that gates skill content and security drift in CI.

## Harnesses and integrations

- [docs/harnesses.md](https://raw.githubusercontent.com/aeonfun/aeon/main/docs/harnesses.md): Advanced harness behavior across the nine supported coding-agent CLIs.
- [harness-adapter/README.md](https://raw.githubusercontent.com/aeonfun/aeon/main/harness-adapter/README.md): The single dispatch path that translates one tool grammar to every harness.
- [apps/mcp-server/README.md](https://raw.githubusercontent.com/aeonfun/aeon/main/apps/mcp-server/README.md): The local stdio MCP server (`bin/add-mcp`): every skill as an `aeon-<slug>` tool in Claude Code or Claude Desktop, run on your machine. Compares it with the hosted server.
- [docs/mcp-oauth.md](https://raw.githubusercontent.com/aeonfun/aeon/main/docs/mcp-oauth.md): Durable dashboard OAuth for OAuth-gated MCP servers.
- [docs/telegram-commands.md](https://raw.githubusercontent.com/aeonfun/aeon/main/docs/telegram-commands.md): The inbound Telegram command surface.

## Reference

- [docs/help.md](https://raw.githubusercontent.com/aeonfun/aeon/main/docs/help.md): The text the Telegram `/help` command sends.
- [docs/ECOSYSTEM.md](https://raw.githubusercontent.com/aeonfun/aeon/main/docs/ECOSYSTEM.md): Ecosystem partners and integrations.
- [CHANGELOG.md](https://raw.githubusercontent.com/aeonfun/aeon/main/CHANGELOG.md): Release history.

## Install as a plugin

Aeon ships the operator console as a plugin for Claude Code and Codex. From version 0.2.0 it also bundles the hosted Aeon Connect MCP connection and a get-started onboarding skill: sign in with GitHub, choose your agent repo, check missing key names and confirm a first heartbeat run. It never reads or writes secret values; setup changes require confirmation.

- Claude Code: `/plugin marketplace add aeonfun/aeon` then `/plugin install aeon@aeon`
- Codex: `codex plugin marketplace add aeonfun/aeon` then `codex plugin add aeon@aeon`
