AI agent shell command policy

Shell commands collapse file, network, credential, and deployment authority into one interface.

Decision boundary

Put the Veto decision before run_shell reaches the system of record. The policy should see actor, tenant, amount or target, environment, and reason before the action executes.

Example protected action

const decision = await veto.protect({
  tool: "run_shell",
  arguments: input,
  context: {
    actorId,
    tenantId,
    environment: "production",
  },
})

if (decision.action === "deny") {
  throw new Error(decision.reason)
}

if (decision.action === "require_approval") {
  await veto.waitForApproval({ decisionId: decision.id })
}

return executeTool(input)

A good page for this search should help the reader implement one governed action, not describe abstract governance. The useful artifact is the action-level decision record.