@sbo3l/autogpt
v1.2.0
Published
AutoGPT command-runner adapter for SBO3L — gate each long-running task step through the policy boundary.
Downloads
127
Maintainers
Readme
@sbo3l/autogpt
SBO3L adapter for the AutoGPT agent framework.
npm i @sbo3l/autogpt @sbo3l/sdkWiring
import { SBO3LClient } from "@sbo3l/sdk";
import { sbo3lCommand, runSbo3lCommand } from "@sbo3l/autogpt";
const sbo3l = new SBO3LClient({ endpoint: "http://localhost:8730" });
const tool = sbo3lCommand({ client: sbo3l });
// Hand `tool.descriptor` to AutoGPT's tool registry.
// Pair each emitted tool call with `runSbo3lCommand` to dispatch through SBO3L.Behaviour
- allow →
{ ok: true, output: PolicyReceipt } - deny / requires_human →
{ ok: false, output: { error, deny_code, audit_event_id } } - transport fail →
{ ok: false, output: { error: "transport.failed", detail } }
The runner never re-throws — every outcome is a structured envelope so the framework's loop can branch and self-correct.
Tests
npm test # 9 vitest passing
npm run typecheck
npm run build # ESM + CJS + d.ts