Name: miroshark-aeon
Author: aaronjmars

Star us ❤️ →

Give it a direction and it ships the work: features, vulnerability disclosures, live apps, deep research - and new skills for itself.

No approval loops. No babysitting. Configure once, forget forever.

This is a live Aeon instance: the growth agent for MiroShark, running the MiroShark repo and $MIROSHARK in public. Want your own? Run now at www.aeon.fun/connect, no clone needed. The framework lives at aeonfun/aeon; start from there, not from this repo.

🤖 Using a coding agent? Point it here: read https://www.aeon.fun/skills/aeon.md and follow the instructions to set up your Aeon agent.

Aeon live demo — the dashboard configures skills, schedules them on cron, and Aeon runs unattended: shipping PRs, disclosing vulnerabilities, and reporting back to your channels.

Quick start

Quick start in four steps: 1 Sign in - sign in with GitHub at www.aeon.fun/connect and create your aeon. 2 Connect a model - paste your Claude subscription token or another key; it is saved as an encrypted secret in your own repo. 3 Pick skills - toggle skills on and set their schedule. 4 Runs itself - Aeon runs unattended on your GitHub Actions.

Make your own agent from the framework repo, aeonfun/aeon, not from this instance. Three ways in:

Browser (recommended): Run now at www.aeon.fun/connect. Sign in with GitHub, create your aeon (a public fork or a private copy of aeonfun/aeon), install the Aeon Connect GitHub App on that one repo, connect a model, pick skills. No clone, no terminal, no Node.
Coding agent: paste read https://www.aeon.fun/skills/aeon.md and follow the instructions to set up your aeon agent into Claude Code, Codex, Hermes or OpenClaw.
Terminal: needs Node.js 20+ and the GitHub CLI (gh auth login).
```
git clone https://github.com/aeonfun/aeon && cd aeon && ./aeon init
```
./aeon init creates your own repo from the template and connects GitHub, a model and a channel. ./aeon opens the same dashboard locally on localhost:5555.

Full guide: www.aeon.fun/docs.

What Aeon can do

A skill is just a file — an Aeon skill is a single SKILL.md: a small frontmatter block (title, category, status) then a plain-language prompt. That one file fans out into real capabilities: launch, deploy, web fetch, and search.

A skill is a Markdown file: some frontmatter, then a prompt. Here's a real one, trimmed:

# skills/digest/SKILL.md
---
name: digest
category: basics                 # which pack it belongs to
description: Generate and send a digest on a configurable topic
requires: [XAI_API_KEY?]         # ? = optional key, bare = required
var: ""                          # per-run input - "solana", "rust", "AI agents"…
mode: write
---

The prompt is the skill. You schedule it, hand it a var, chain it into others, and Haiku rates every run. How packs work: docs/skill-packs.md.

Six skill packs, 82 skills total: Core (fleet coordination, self-config, liveness), Evolution (authors and heals its own skills), Basics (simple runnable skills), Dev & Code, Crypto & Markets, and Productivity.

Full catalog - all 82 skills by pack →

Community skill packs →

Support seven harnesses: Claude, Grok, Codex, Pi, Vibe, Kimi, fx

Seven engines, one socket - a SKILL.md flows through run-harness into any of seven agent CLIs: Claude, Grok, Codex, Pi, Vibe, Kimi, and fx. Every harness honors the same contract: result, usage, session.

The same SKILL.md runs on any of seven agent CLIs - Claude, Grok, Codex, Pi, Vibe, Kimi, fx - behind one run-harness contract (same result, usage, and session shape). Swap the harness; nothing else changes. How the contract works: docs/harnesses.md.

Why "the most autonomous"

Most agent tools keep you in the loop - approve this call, review this diff. Aeon is built for the work you want done while you're away, and it's the only framework that does all four unattended: runs on a schedule, remembers across runs, reacts to conditions, and repairs its own broken skills. The most autonomous agent is the one that never asks.

Full comparison vs Claude Code, Hermes, and OpenClaw: SHOWCASE.md.

Autonomy spectrum

Proof of work

Proof of work, live at aeon.fun: 2M GitHub stars secured across 69 open-source repos, 68 products and agents built on Aeon, and 12 community skill packs.

Live at aeon.fun: 2M GitHub stars secured - real vulnerabilities found, patched, and responsibly disclosed across 69 open-source repos (Alibaba, Tencent, Vercel, and more), many rated High or Critical. 68 products built on top of Aeon.

Every disclosure → · ecosystem · community packs

It ships real work

It ships real work — four things Aeon does end to end: ships code (PRs to your repos), deploys apps (live to Vercel), finds bugs (real vulnerabilities, privately disclosed), and writes skills (new capabilities from a sentence).

feature ships code to your repos, deploy-prototype ships live apps to Vercel, vuln-scanner finds and privately discloses real vulnerabilities, create-skill writes new skills from a sentence. How each works: CORE.md.

It heals itself

Anatomy of a skill run

A model scores every run 1–5; heartbeat → skill-health → skill-repair → self-improve detect and fix broken skills without you, and aeon-doctor lints the config itself. How the loop closes: CORE.md.

It replicates

It spawns itself — one Aeon forks into a fleet of specialized instances (crypto, security, research), each its own node with isolated billing and no propagated secrets.

spawn-instance forks Aeon into a new specialized instance (var: "crypto-tracker: monitor DeFi protocols"), picks relevant skills, and registers it - no secrets propagated, billing isolated. fleet-control health-checks and dispatches across the fleet.

Guardrails

Autonomy needs brakes — four safety brakes on by default or one flag away: read-only skills can't touch the repo, irreversible actions fail closed, an optional auth gate approves every run, and secrets stay off the command line.

Read-only skills can't touch the repo, irreversible actions fail closed, an optional Fleet Watcher gates every run, and secrets stay off the command line. Details: Configuration.

Configure

Everything lives in aeon.yml - schedules (standard UTC cron), the per-skill var input, model, auth, notification channels, and API keys:

skills:
  digest:
    enabled: true
    schedule: "0 14 * * *"    # daily 2pm UTC
    var: "solana"             # per-run input

Full reference - scheduling, var, models, authentication, notification channels, API keys, guardrails: Configuration.

GitHub permissions: the built-in GITHUB_TOKEN covers same-repo work; add one classic PAT as GH_GLOBAL with repo + workflow scopes to drive every cross-repo, private, disclosure, and workflow-editing skill. No read:org / admin:org. See Cross-repo access.

Integrate Aeon

An Aeon instance is just a GitHub repo + Actions, so GitHub's API is Aeon's API - one GitHub App drives your users' instances (dispatch skills, edit config, write sealed secrets) with no PATs or LLM billing on your side.

Full walkthrough - App setup, tenant isolation, driving skills, shipping your own pack: ADK - Aeon Developer Kit.