agentproto steward: agentproto

agentproto steward [--apply] [--idle <min>] [--min-confidence <x>]
                   [--judge <auto|jev|agent>] [--ask-sessions] [--wait] [--json]

Convenience shortcut over agentproto workflow run-file for the built-in session-steward app's workflow: wrap up idle agent sessions. Needs a running daemon (agentproto serve) — the workflow reads live sessions and its agent judge spawns real (one-shot) sessions. A dry run unless --apply.

One run:

Plans with session_wrapup_plan: every idle agent session is close, stuck, judge or keep.
With --apply: closes close-class sessions as done and stuck-class ones as abandoned — resumable, with a recorded outcome.
Collects compact, read-only evidence (session_evidence) for up to 15 judge-class sessions, most RAM first.
Judges each one: with Jev (TypeSafe System One — a calibrated choice with probabilities) when jev.apiKey from ~/.agentproto/config.json or the JEV_API_KEY env var resolves, else a one-shot agent judge (sonnet). A Jev failure falls back to the agent judge for that session; a malformed or failed judgement is active and never acted on.
With --ask-sessions: asks low-confidence idle sessions directly whether they're done (one prompt each, ~3 min bounded wait).
With --apply: confident (≥ --min-confidence) done/abandoned verdicts close the session; blocked/needs-input only flag it.
Reports a markdown table plus RAM freed / still held.

keep-class sessions and the calling session (AGENTPROTO_SESSION_ID) are never touched, and session_wrapup_apply re-checks every id right before acting.

Flag	Default	Description
`--apply`	`false`	Close / flag sessions. Without it: plan + verdicts only, nothing is touched.
`--idle <min>`	`30`	Idle threshold in minutes.
`--min-confidence <x>`	`0.8`	Judge confidence (0..1) needed to act on a verdict.
`--judge <backend>`	`auto`	`auto` (Jev when `jev.apiKey` config or `JEV_API_KEY` env resolves, else the agent judge), `jev`, or `agent`.
`--ask-sessions`	`false`	Ask low-confidence sessions directly — spends a turn in their conversation.
`--wait`	`false`	Block until the run ends, then print its markdown report. Exit `0` when done, `1` when it failed or was cancelled.
`--json`	`false`	Print the raw `workflow_run_file` reply (with `--wait`: the finished run record).

This starts the run and returns immediately — poll it with:

agentproto workflow status <runId>

Or pass --wait to block until it finishes and print the report inline.

Examples

# Dry run: plan + verdicts, print the report
agentproto steward --wait

# Close / flag confident verdicts
agentproto steward --apply --wait

# Stricter: only sessions idle an hour, agent judge, 0.9 confidence
agentproto steward --apply --idle 60 --judge agent --min-confidence 0.9

Scheduling it

packages/apps/session-steward/routines/session-steward-hourly is an AIP-41 ROUTINE.md template — hourly, apply: true, askSessions: false — shipped enabled: false, so nothing starts closing sessions on install. Its own doc lists the enabling steps.

agentproto steward

ExamplesCopy Anchor Link

Scheduling itCopy Anchor Link

On this page

Examples

Scheduling it