definition.status

Read one Status snapshot of a Loop from disk — no Run started, no Lock claimed.

status() returns one snapshot of what has happened, read from disk. It starts no Run and claims no Lock.

import { Loop } from "@loop.js/core"

const loop = Loop.define({ goal: { file: "GOAL.md" } })
const status = await loop.status()

if (status.running) console.log(`pid ${status.pid} is on round ${status.round}`)

LoopStatus

type LoopStatus = {
  running: boolean
  pid?: number
  round: number
  usd: number
  lastExit: Exit | null
  verdicts: ({ round: number; ok: boolean; impossible: boolean; reason: string })[]
}

Field	Meaning
running	Whether a Run currently holds the Lock.
pid	The Lock owner’s process id — present iff running.
round	The resume cursor — where the next Run picks up.
usd	Total spend across the whole Loop so far.
lastExit	How the last Run ended, or null if none has.
verdicts	The full Verdict history with reasons, one entry per judged Round. impossible is always present on the wire — false when ok.

Reading lastExit:

• { settled: true, verdict } — a Verdict ended the Loop. verdict.ok says which settle it was: success, or give-up (ok: false with impossible: true).
• { settled: false, cause, reason } — the Run was interrupted (budget | rounds | cancel | error | yield) and the Loop stays live for the next Trigger.

A never-run Loop is not an error — it reads as the zero state.

Agent runs keep no Record

Agent.define(config).run() runs the Execute phase once, ungraded: no Verify, no Verdict, no Lock, no Record. Because it keeps no Record there is nothing to read back — AgentDefinition has no status().

import { Agent } from "@loop.js/core"

const agent = Agent.define({ goal: "Summarize the repo layout" })
const exit = await agent.run().done()
// { finished: true, reason }  — reached a terminal stop, not a "done" claim
// { finished: false, cause: "budget" | "cancel" | "error", reason }