Let agents list and message a project's agent definitions through the t3-code MCP server

[fadymondy]

Problem

Many repos define specialised agents as files (for example .claude/agents/*.md: a reviewer, a QA agent, a docs writer). Inside T3 they're only usable through Claude's own Agent tool, as nested subagents invisible to T3. Other providers can't use them at all, and T3's thread list and lineage don't show them.

Proposal

Add two tools to the t3-code MCP server:

• agent_list: the project's agent definitions (name, description, model, source file).
• agent_message: send a message to an agent by name. The first message launches an ordinary top-level T3 thread titled @<name>, with the definition's instructions as its system prompt and the definition's model or the caller's. Later messages go to the same thread.

Everything runs through the existing t3_thread_launch / t3_thread_send paths and the caller's permissions, so agents show up in the sidebar, keep their history, and work from any provider.

Out of scope

An agent picker in the composer, dynamic model switching and editing agents. (#17924 proposes a much broader version; this is a narrow first step towards it.)

Status

Working in a fork, with tests. I'd like maintainer approval of this scope before opening a PR.