Observed facts: OpenRig defines agent teams in YAML through RigSpec, including pods, edges, and continuity policies. Observed facts: it can launch and manage Claude Code and Codex sessions in tmux, expose their topology in a TUI, and provide rig send, rig broadcast, and rig chatroom for coordination. The mechanism is a local control layer that gives separately running coding agents stable identities, a shared topology, readiness checks, messaging paths, and snapshot-and-restore operations. That makes the unit of work a recoverable team configuration rather than an ephemeral prompt session.
Why now: coding-agent usage is moving from a single interactive assistant toward small, specialized teams where review, delegation, and context continuity create practical coordination costs. The first implication is that engineering teams can standardize multi-agent workflows around explicit roles and review paths instead of relying on ad hoc terminal discipline. A second implication is that the harness becomes part of the workstation’s security boundary: OpenRig documents writes to Codex and Claude configuration, hooks, trust records, and workspace settings. Evidence boundary: these capabilities and configuration effects are described by the project’s own repository; this source does not establish reliability, security quality, or productivity gains in independent deployments. The first watch condition is whether early users report that topology recovery and cross-harness handoffs reduce coordination overhead without creating unacceptable configuration or permission risk.
This starts a distinct analytical record because no prior New Runtime record was supplied, and the project concerns a specific local control plane for mixed Claude Code and Codex teams.