@guyghost/swarm-dao-graph

Swarm DAO Graph Engineering run executor.

Runs the frozen Graph Engineering machine (in @guyghost/swarm-dao-core) as journal-replayed runs on disk: every submitted signal is validated (AI signals never carry state targets, commands, approvals, retries, cancellations, or permission decisions), journaled to journal.ndjson, and the snapshot is persisted for deterministic replay.

Usage

import { createGraphRunner, validateGraphSignal } from "@guyghost/swarm-dao-graph";

const runner = await createGraphRunner({ evidenceRoot: ".dao/graph-runs", runId: "my-run" });
const result = await runner.submit(signal);

Or through the CLI entry:

import { runGraphCli } from "@guyghost/swarm-dao-graph";
const code = await runGraphCli(["status", "--run-id", "my-run"]); // 0 | 1 | 2

The swarm-dao CLI exposes the same surface as swarm-dao graph <init|status|submit|implement>; exit codes are 0 (success), 2 (machine rejection), 1 (usage or execution error). graph implement prepends CLASSIFIER_CHARTER and routes the implementer through evaluateAttempt before submitting an AI artifact (IMPLEMENTATION_READY / IMPLEMENTATION_FAILED).