OpenRig explained: coordinate coding agents without losing the operator
OpenRig security operations: hooks, trust and workload scope
Audit machine writes and agent permissions before using a real repository
What you will learn
- Back up the files that change
- Separate command allowances
- Inspect data flows
Before you start
- Node 22 or 24 and tmux on a supported host
- One working provider account and a disposable repository
Turn the first owner/checker run into evidence for a safe rollout decision
Key takeaways
- Setup preview does not include every later startup write.
- Command allowances require an explicit user choice.
- Hook payload claims do not cover provider data flows.
Back up the files that change
The README lists `.tmux.conf`, daemon state, Codex `config.toml`, Claude settings and project-local hooks among possible writes. It also says startup may pre-trust a workspace.
A dry-run of `rig setup` does not cover every later daemon or seat side effect. Compare the file inventory before and after daemon start and first rig launch; keep the backup outside the test repository.
Separate command allowances
The built-in bootstrap does not grant blanket `rig` command allowances. Agent-guided setup asks the user whether to add native rules at a chosen scope; No leaves those settings unchanged.
The project also notes selected resources can change MCP settings and Claude permission mode. Review those resources explicitly. Do not interpret a consent to run `rig` as permission to change arbitrary repository code.
Inspect data flows
Activity hooks send event type, seat and runtime identity, timestamps and native session identity to the configured daemon; the documented payload excludes prompt text and tool arguments. Context collection stores usage and transcript-path metadata locally.
Provider and MCP integrations have their own data flows. Keep task content synthetic during a pilot and verify log retention, network exposure and token handling against your organization’s policy. This article is not a security audit.
Decision guide
| Criterion | Option A | Option B |
|---|---|---|
| Best when | You need predictable behavior and easy auditing | You need adaptive optimization and have reliable telemetry |
| Main risk | May leave performance on the table | Can become difficult to explain or debug |
Implementation steps
- 1
Back up provider and workspace files before first launch.
- 2
Inspect consent scope, trust changes and selected MCP resources.
- 3
Use synthetic tasks while tracing activity and context metadata.
Copy-ready example
setup preview -> daemon hooks -> seat trust
activity relay -> daemon metadata
provider / MCP -> separate data flowsFrequently asked questions
Does OpenRig automatically approve every `rig` command?
The pinned README says built-in bootstrap does not; guided setup asks for an explicit choice.
Is changing OPENRIG_HOME enough to sandbox hooks?
No. Provider configuration and workspace files can be outside that directory.
Sources
- OpenRig / README.mdSource checked 2026-10-04
- OpenRig / docs/reference/scoped-operating-posture.mdSource checked 2026-10-04
- OpenRig / docs/reference/telemetry.mdSource checked 2026-10-04
- OpenRig / docs/reference/host-resources.mdSource checked 2026-10-04
- OpenRig / packages/cli/src/commands/setup.tsSource checked 2026-10-04