tom2698

AgentPortal — AI skill for Claude Code

AI community

Code prototype of a control room for a team of AI agents: Slack-style chat, agent-to-agent @mentions, schedules, web + Expo app, and a runner that drives Claude Code.

How to install AgentPortal

This entry records only its repository, not the path inside it, so there is no exact command to give. Open tom2698/AgentPortal and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What AgentPortal does

Code prototype of a control room for a team of AI agents: Slack-style chat, agent-to-agent @mentions, schedules, web + Expo app, and a runner that drives Claude Code. The production version is private.

Alternatives in AI

README

Agent Portal

A control room for a small team of AI agents. You chat with them in Slack-style channels and DMs, they hand work to each other with `@mentions`, run on schedules, and report back. There's a web portal and a phone app, both driving the same API.

**This is a code prototype.** I built and run the production version for my own business, and that codebase is private. This repo is a trimmed-down, self-contained copy of the same design: the real UI, the real queueing and hand-off logic, and the real runner, with an in-memory store and a stand-in "demo brain" so it runs anywhere in a couple of minutes.

Web: team chat Web: dashboard (dark)
Team chat Dashboard
Phone: dashboard Phone: conversation Phone: agent settings
Phone dashboard Phone conversation Phone agent

More in [`docs/screenshots`](docs/screenshots).

What it does

  • Team chat. #general, #standup and a DM with every agent. Tag @Sage to wake Sage; @everyone wakes the team.
  • Agents talk to each other. An agent's reply can @tag a teammate, which queues that teammate with the conversation so far. Threads are capped (8 back-and-forths) and say so when they pause, so two agents can't loop forever.
  • Agents can tag you. @Tom in an agent's message pings the owner's phone; so does a reply in your DM or an answer to something you asked. Agent-to-agent chatter doesn't.
  • Schedules. Each agent can have a cron schedule ("Every day at 4:00am") with a one-line summary. Scheduled work reports to #standup.
  • Per-agent brain and model. Claude Code or Hermes, and which Claude model (Fable, Opus, Sonnet, Haiku, or any name).
  • Per-agent permissions. An explicit tool allowlist, or "full access" with a deny list that still