Permissions

Permissions gate tool calls before execution. Every pending tool invocation is evaluated against a PermissionPolicy that returns a Decision: allow, deny, or require human approval.

The runner integrates policy with guardrails (middleware docs) and durable checkpoints so channel bots and desktop hosts can pause a run, surface an approval UI, and resume with runAgentResume.

Decision shape

interface Decision {
  allowed: boolean;
  requires_approval: boolean;
  matched_rule_id?: string;
  reason?: string;
}

Quick start

import { Maniac, StaticPermissionPolicy } from "@maniac-ai/agents";

const policy = new StaticPermissionPolicy(
  [{
    id: "approve-chat-writes",
    principal: "*",
    scope: { toolset: "chat", arg_constraints: [] },
    effect: "require_approval"
  }],
  { allowed: true, requires_approval: false } // default: allow everything else
);

const app = new Maniac({ model, memory, policy });

Set policy on the Maniac constructor for app-wide defaults, or on individual Agent specs to override per agent.

Topics

• Policies — PermissionPolicy, StaticPermissionPolicy, rule effects, path constraints
• HITL & checkpoints — pause/resume, RunCheckpoint, SessionPermissionCache, RunCheckpointStore

API reference

Generated TypeDoc: permissions module.