Skip to content

AGENTS.md

Latest commit

 

History

History
44 lines (32 loc) · 1.68 KB

AGENTS.md

File metadata and controls

44 lines (32 loc) · 1.68 KB
Raw

Automux

Multi-agent orchestration through tmux. Bun + TypeScript.

Architecture

Human → automux start → Main agent (pane 0) → automux spawn → Subagents (right panes)
                              ↓                                      ↓
                         uses automux CLI                     uses files only
                         merges, rebuilds                reads brief, writes output
  • tmux — visual layer. Human watches, agents work in panes.
  • Files — communication. Briefs in, output files out. No pane scraping.
  • SQLite — plumbing. Pane IDs, session IDs, status tracking.
  • Git worktrees — isolation. Each subagent gets its own branch.
  • automux CLI — lifecycle. The main agent's only interface to tmux.

Dev Commands

bun install          # deps
bun run dev          # run from source
bun run build        # compile to dist/ (updates global automux)
bun run typecheck    # type check
bun test             # tests

Rebuild after source edits: ~/.bun/bin/automux → dist/cli/index.js.

Self-Development

You are the main agent. You develop automux using automux.

For the full orchestration guide, read AUTOMUX.md. For commands, use /automux-cli. For the workflow loop and recovery patterns, use /automux-workflow.

Rules

  1. Use automux commands for all tmux operations — never raw tmux
  2. Write the brief before spawning — never spawn without context
  3. Read output files for results — pane capture is the fallback, not the default
  4. After merging source changes, always bun run build before using new features
  5. Subagents communicate through files only — they don't know about tmux or automux