Agent Skills (Open Standard)
SDK / ToolThe open Agent Skills standard — SKILL.md folders any agent can load, governed by the Agentic AI Foundation since 2026-10-22.
- official
- spec
- agent-skills
- standard
A self-hosted dispatcher for Claude Code and Codex sessions: run 100 like 5 — the hub keeps them in order, hands out the work and brings you only the decisions, in your browser or in Telegram
Hub by Icons8 is a self-hosted dispatcher for Claude Code and Codex sessions: it drives them (it brings its own copy of the program; you sign in to your Claude account, or to Codex) with no prompt per action — treat it like a shell on your machine — and it runs on your computer or your own server. You talk to one coordinator session in plain words; it decides who does what, starts the right session and brings the result back. Permanent agents have their own area, tools and notes; short-lived executors work one task in their own git worktree and go to sleep with their conversation intact. Every session lands in one feed, and the Now tab shows only what waits for you, what is new and what is working; questions arrive as cards with buttons on the page and in Telegram, and a reply goes back to the session that asked. A message to a busy session waits for the end of its turn, an urgent one gets through mid-turn, and delivery is visible. An event journal is the single source of truth and the state is a pure function of it; a watchdog notices loops, silent sessions and sessions that repeat the same error, and a fallen session restarts with its conversation. Extending needs no code changes: skill sets, MCP servers, your own processes, page tabs and themes live in your own folder and survive updates. Secrets are sealed (AES-256-GCM, the key outside the hub folder — Keychain, DPAPI, or a file on Linux) and used with hub secret fill / hub secret exec so agents never see the value. One machine can run several hubs, and hubs of different people can be linked by encrypted invitation through an optional relay that passes messages without seeing the text. One-line installer for macOS, Linux and Windows. Experimental, and not affiliated with Anthropic.
Use when Claude Code or Codex work has outgrown one terminal — ten parallel sessions across repositories, a coordinator splitting a feature into executor tasks, or overnight runs you want to steer from your phone in Telegram. Skip it if you only ever run one session at a time: it adds a whole service that is itself something to manage.
One line in a terminal: curl -fsSL https://icons8.github.io/hub/install.sh | bash # macOS 13.5+, Linux (glibc 2.28+) irm https://icons8.github.io/hub/install.ps1 | iex # Windows 10/11, in PowerShell Nothing has to be installed first and no administrator rights are needed: the line puts the hub's environment (Python, Node, gh, ffmpeg; git too on macOS and Windows), its service and the hub command into your user folder, installs Claude Code and a browser for agents, then opens the page at http://127.0.0.1:8787/. What is left for you is Sign in to Claude on the band above the page — or Sign in to Codex, which restarts the hub on Codex and installs its CLI. For a cloud server, paste the same line with -s -- --server --yes added into the provider's user-data field (the README's table marks Hetzner, DigitalOcean and AWS EC2 as tested), then reach the page through the SSH tunnel it prints.
Experimental, and the README's first block says so plainly: the hub drives sessions with no prompt per action (sessions run in bypassPermissions mode; on Codex: never ask, no sandbox), it is not a sandbox and not a replacement for Claude Code or Codex — read SECURITY.md before running it anywhere that matters. The page and its Files view have no sign-in unless you set a password: anyone who can open 127.0.0.1:8787 on the same machine acts as the operator and sees the whole disk as the hub's user, so install only where you are the only user or set a password for every hub. Cloud servers read the disk including the Claude sign-in; the README says to run sudo -iu hub claude auth logout before deleting a machine. Pull requests are not accepted for now — a deliberate decision of the small team; bug reports and Discussions are welcome, extensions are your own modules and plugins. What leaves the machine is documented in the security table (update checks, the plugin catalogs it reads daily, the downloads it performs), and there is no telemetry unless you switch one on. Apache-2.0. 23 stars as of writing (2026-11-20).
Repository
icons8/hubCharted
The open Agent Skills standard — SKILL.md folders any agent can load, governed by the Agentic AI Foundation since 2026-10-22.
A local launcher that points Claude Code and seven other coding harnesses at any OpenAI-, Anthropic- or Responses-compatible endpoint — without editing a single config file.
Answers macOS Spotlight and Siri prompts with your signed-in Claude Code account — an experimental macOS 27 model-delegation provider.