OpenCode workers limited to approved Go models
Rig now restricts OpenCode workers to an approved list of Go models, so routing stays predictable.
Rig is an open-source CLI harness for coding agents. You talk to one parent agent; it scopes the work, briefs worker agents over MCP and checks their results against your requirements. Done means accepted checks, not exit 0.
Talk to a single parent agent. It briefs workers over MCP and keeps them on task.
Eligible work splits into a graph of up to 12 nodes, with parallel workers on separate files.
Work is checked against declared requirements and optional acceptance contracts before it counts.
Tasks go to the eligible worker best suited for them.
rig tuiA terminal UI with Jobs, Queue and Workflows tabs.
Safe updates with rollback, a task-aware doctor, and opt-in computer-use and BrowserSkill backends.
.rig/harness.tomlrig off and rig onRig is an open-source CLI harness for coding agents from Brasth. One parent agent scopes the work, briefs worker agents over MCP and verifies their results.
Rig supports Codex, Claude Code, Grok, OpenCode, Cursor and others as parent or worker agents.
A job counts as done only after the parent agent accepts it against the declared requirements and any optional acceptance contract, not just because a process exited with code 0.
After rig init and configuring workers in .rig/harness.toml, type normal prompts into your parent agent. The project recommends not using rig run for normal work.
Yes. Rig is MIT licensed. The source is at github.com/Brasth/Rig.
Rig now restricts OpenCode workers to an approved list of Go models, so routing stays predictable.
Review receipts are validated against the job that wrote the code before the parent accepts the handoff.
Fixes how a finished workflow graph hands its results back to the parent agent.
Groundwork for running workers in separate git worktrees, plus refreshed layouts for rig tui.