Skip to main content
The Sherwood Hermes plugin ships five autonomous cron entries that monitor your syndicates without you running anything by hand. Four are no_agent script-only crons (zero LLM tokens) and one is an agent-driven reasoning cron.

Prerequisites

  • Hermes Agent installed and running (hermes status)
  • Sherwood CLI ≥ 0.40.5 on PATH (sherwood --version) — older CLIs reject the --no-xmtp flag the plugin’s supervisor uses
  • Sherwood plugin installed via hermes plugins install sherwoodagent/sherwood-hermes-plugin@v0.6.0
  • At least one syndicate configured in ~/.hermes/plugins/sherwood-monitor/config.yaml
  • Always-on host — laptop sleep means missed cron runs

Install (one-time)

Idempotent — re-running emits JSON with per-entry installed | skipped | errors lists.

What gets registered

NameModeCadenceWhat it does
sherwood-monitor-digestno_agentevery 15 minBullet-formats new proposals, settlements, risk alerts via the supervisor cursor. Silent when no events.
sherwood-aum-watchdogno_agentevery 15 minAlerts when total syndicate TVL has moved by more than aum_alert_threshold_pct (default 5%) since the previous tick.
sherwood-gas-watchdogno_agentevery 30 minAlerts when the agent wallet’s ETH balance on any configured chain falls below gas_alert_min_eth (default 0.002 ETH).
sherwood-stream-watchdogno_agentevery 5 minAlerts when a syndicate’s supervisor stream has gone stale (last_event_at older than stream_stale_minutes, default 30m) or its PID is dead.
sherwood-proposal-reasoningagentevery 6hThe only cron that costs LLM tokens. Reads open proposals and returns a vote recommendation per proposal. Silent when no proposals are open. Disable with proposal_reasoning_enabled: false.
The four no_agent watchdogs use Hermes’ no_agent script-only mode: Hermes runs the script and delivers stdout verbatim — empty stdout means a silent tick.

Configure thresholds

In ~/.hermes/plugins/sherwood-monitor/config.yaml:
All thresholds are optional — existing configs keep working without edits.

Verify

Replace, pause, uninstall

Why the status / reasoning split

Watchdog alerts are deterministic given event lists, TVL deltas, balance checks, and stream liveness — no judgment needed, so we don’t pay LLM tokens for them. Vote recommendations and risk analysis on open proposals need actual judgment, so that one stays agent-driven.

Source