Agents section: projects you chat with, plus a few tools so they know about each other

[McCalebTheSecond]

Rough idea, not a spec. Posting to see if the direction fits.

Idea: an "agent" is just a regular T3 project shown a little differently. Mark a project Show in Agents and it gets a row in a small Agents section at the top of the sidebar. Its threads stay normal threads: use one or many, and Working, Settled, snooze, scheduled tasks and delegation work as they do today. Think a very basic, light OpenClaw or Hermes, built from what T3 already has.

UI

AGENTS                        +
 [O] Ops           ● 2 working
 [R] Research      ● needs you
 [W] Writer          3h
──────────────────────────────
Pinned / Active / Working / Settled   (unchanged)

• Each row shows the project's icon and name, plus a status rolled up from its threads.
• Click an agent to get a composer ("Message Ops…") that starts a new thread in its project, with its recent threads below.
• Hover + starts a new thread with that agent; its menu can open the latest thread or filter the sidebar to it.
• Name, icon, instructions (AGENTS.md), default model and settings all come from the project, so there's nothing new to configure.

A few built-in tools for threads in agent projects, added to T3's MCP server:

• agents_list: the other agents, with a short description and current status.
• agent_message: message another agent. Starts or continues a thread in its project through the existing thread launch/send paths.
• Shared memory: a small notes store every agent can read and append to.
• Shared skills: one skills folder T3 exposes to every agent's provider, so a skill added once works for all of them.

Related

• Agent-first view: one continuous conversation per project, sessions hidden #24351: one continuous conversation per project. Same starting point; this keeps normal threads and the normal lifecycle.
• feat(home): add Home, an agent that manages threads across your machines #15975: Home, one built-in agent at the top of the sidebar. This would let any project be one.
• Let agents list and message a project's agent definitions through the t3-code MCP server #18131: agent_list / agent_message for agent definition files.
• Skill manager: view, add, edit, and link skills across every harness #15823, [Feature] Manage skills and agents from T3, route to providers via symlinks #7334: skill management across harnesses. This would only need a shared folder, not a full manager.