Back to Ecosystem Pulse
ToolProduction Ready
openrig
by mvschwarz
CLI harness to run Claude Code and Codex together for multi-agent coding experiments
TypeScript
Updated Oct 7, 2026
Share:
Summary
Orchestrates multiple coding agents (Claude Code and Codex) in a single TypeScript harness for side-by-side workflows orchestrator–worker pattern. Uses a CLI + tmux-driven runner to coordinate agents, capture outputs, and let them collaborate or compete on coding tasks. Includes utilities for piping prompts, logging interactions, and replaying sessions for analysis, guided by the supervisor pattern supervisor pattern.
Key Benefits
As teams compare model behaviors for coding tasks, side-by-side multi-agent runs surface differences in reliability, failure modes, and solution styles. A reproducible harness like this makes it practical to collect agent interaction logs and build an agent track record for downstream evaluation. That visibility is essential when reasoning about agent-to-agent evaluation and forming reputation signals across models agent-to-agent evaluation.
Ideal For
Researchers and engineers running comparative or collaborative experiments between coding-focused LLMs and collecting interaction logs for evaluation, particularly in contexts that benefit from event-driven agent pattern guidance.
Real-World Examples
- Compare outputs and failure modes of Claude Code vs Codex on the same coding tasks
- Record and replay multi-agent coding sessions for evaluation or benchmarking
- Orchestrate collaborative/competitive agent workflows to study delegation and interaction patterns
Works With
openaianthropictmuxtypescript
Topics
agent-harnessagent-orchestrationagent-skillsai-codingclaude-codeclicodex-climulti-agentmulti-agent-systemstmux+1 more
Similar Tools
autogenagent-playground
Keywords
multi-agent orchestrationmulti-agent trustagent-harnessclaude-codecodex