Orbit Authority Guardian

Guardian keeps AI agents from crossing lines you did not mean to move.

Wrap your agent tools — shell commands, repo changes, deploys, secrets access, and cloud actions — with policy checks before they run.

Orbit Authority cockpit

Guardian action trace

Tool calls, including bypass attempts, become authority requests before risky commands touch the machine.

Sample queue view
policy: no-prod-writes · Production changes need explicit approval.policy: spend-limit-5k · Over $5k needs approval.policy: human-approval-required · A named owner must approve.policy: standing-refunds-500 · Refunds under $500 can proceed.policy: secret-access-blocked · Secret paths are blocked.
ActorRequested actionSurfaceRiskVerdict

Click any requested action, or press Enter or Space while focused, to inspect its policy and receipt details.

Built for real agent toolchains.

Guardian wraps the tools your agents already use.

  • Hermes

    Cron jobs, PR watchers, release scripts, long-running automations.

  • OpenClaw

    Local agent runtime and tool-call checkpoints.

  • Claude Code

    Hooks around shell, file, repo, and package-manager actions.

  • Codex

    Wrapper checkpoints for command and repo mutation paths.

  • Wrappers

    gh, kubectl, terraform, docker, ssh, and internal CLIs.

  • Custom hooks

    Any surface that can ask: actor, action, scope, context.

The authority loop

  1. Actor requests a consequential action.

    agent: ORBIT_BYPASS=1 terraform destroy workspace=prod

  2. Orbit Authority checks authority and scope.

    policy: production protected + bypass attempt detected

  3. Allow, escalate to a human, or deny.

    Blocked before it runs.

  4. Receipt and audit trail recorded.

    A signed receipt records who, what, which rule, the decision, and when.

How Guardian governs

Approval first where enforcement is real. Honest labels everywhere else.

Protected action pauses

Guardian pauses risky commands — sudo, SSH, cloud changes, GitHub admin, Terraform/Kubernetes, deploys, destructive DB actions — until you approve.

Verified approver decides

ORBIT asks the configured approver and returns allow, deny, time-bound approval, or escalation timeout based on policy.

Fails closed when protection is active

When Guardian is protecting an action, an unreachable gateway or no approver response blocks the action rather than silently allowing it.

Watch mode is labeled

Where Guardian can only watch, it says so.

Proof without leaking the work.

Receipt-backed decisions

Every governed decision should produce an auditable receipt with actor, action hash, policy reason, verdict, approver path, timestamp, and key reference.

Redacted evidence

Receipts prove the decision without exposing secrets, raw prompts, raw file contents, or private infrastructure details.

Every tool call has a policy now.

ActionVerdictReason
sudoESCALATEOwner approval required.
rm -rf outside temp pathsBLOCKDestructive filesystem action.
git push --forceESCALATEHistory rewrite needs explicit approval.
terraform destroyBLOCKShared infra protected by policy.
ORBIT_BYPASS=1 secret file readBLOCKBypass attempt and secret access blocked by policy.
npm testALLOWRead-only repo-local command.

Wrap the places where agent decisions become real

  • OpenClaw / Hermes · Agent runtimes and delegated work
  • Claude Code / Codex · CLI hooks before tool execution
  • Stripe · Refund, payout, charge, and plan calls
  • Broker path · Orders before execution
  • App mutations · Customer-data and account changes
  • Deploy tools · gh, kubectl, terraform, CI/CD