ParallelCodingAgentsTmux
title: Parallel Coding Agents with tmux radar_quadrant: Techniques radar_ring: Assess radar_position: inner
Parallel Coding Agents with tmux is a practitioner technique for running multiple Claude Code agents concurrently using terminal multiplexing. Each agent runs in its own tmux pane, given an isolated Markdown spec file describing its sub-task. Coordination happens through shared output files read when agents complete, requiring no special framework or orchestration layer.
Claude Code is single-threaded by design: one agent, one conversation, one task at a time. Teams working on large codebases often need multiple agents working on independent sub-tasks simultaneously — different modules, different tests, different services. The typical workaround is sequential execution, which is slow.
This technique fills the gap: open tmux, create one pane per agent, hand each agent a Markdown spec scoped to its work, and let all panes run in parallel. Agents stay isolated from each other and communicate only through files the orchestrating developer reads when each finishes. The approach requires disciplined spec writing — each Markdown file must be self-contained — but adds no dependency on emerging orchestration tools like cord.
Placed in Assess at inner position because it is a proven, low-overhead approach that works today with standard tools available to any Claude Code user. The inner position reflects immediate applicability: any developer running Claude Code on a project large enough to decompose into parallel workstreams should evaluate this pattern. It occupies the practical near-term end of a spectrum that runs through cord toward full multi-agent orchestration frameworks.
