shagghiesuperstar/hermes-fleet-management
Manage any Nous Research Hermes Agent fleet from Grok Bot. Golden Rule: read the official Hermes docs before any action (built-in hermes_docs tool). Fail-loud health, version and skill drift, async ask via the native Runs API, inference status, dry-run-by-default gateway restart and canary updates, mailbox fallback, and a compound-learnings loop.
Use when deciding how Grok Bot should talk to a Hermes agent - HTTP API primary, file mailbox fallback - or when an ask fails and a fallback is tempting.
Use after any Hermes incident, surprise, or fix. Turn the lesson into a skill or rule plus a regression check, cite the official doc it was checked against (or mark it unverified), and log it in CHANGELOG and docs/LESSONS.md.
Use when hosts that should be identical behave differently. Detect version, skill and config drift across the fleet and converge it without blind overwrites.
Use when a deep troubleshooting dig would burn the controller's tokens. Delegate it to a Hermes session on a healthy host through ask/get_result, and keep the controller thin.
Use when a Hermes host is Desktop-primary (a workstation running Hermes Desktop with no supervised messaging gateway), so fleet tools don't break it by installing or respawning a gateway.
Use for daily or on-demand Hermes fleet health checks. Fail loud - every configured host is reported, unreachable hosts are FAIL never silently skipped; stay quiet only when fully green.
Use when a Hermes gateway or Desktop is down, crash-looping, or unstable on one host, or when a restart is requested. Covers read-only triage, attribution, and a safe restart through the dry-run and confirm flow.
Use on first install of Hermes Fleet Management, or when the user asks how to connect Grok Bot to their Hermes agents. Walks through enabling the Hermes API server, configuring hosts and keys, and a first smoke test.
Use before ANY Hermes Agent action, answer, command, config change, or troubleshooting step. Read the official Hermes docs for that topic first and never guess. Every other skill in this plugin runs this as step 0.
Use when choosing providers or models by cost and task, checking quota or headroom before long runs, or when a provider is exhausted and you need a fallback chain.
Use for routine Hermes host hygiene and after any sprint. Inspect logs, disk and checkpoints, clean orphan processes, and confirm the talk path is clean before declaring DONE.
Use before and after OS updates or reboots on Hermes hosts. Make the network overlay and Hermes gateway auto-start, and know the full-disk-encryption and auto-login caveats that keep a headless host offline.
Use when managing Hermes harness identity across hosts: profiles, SOUL.md and context files, and keeping skills in sync fleet-wide.
Use when a Hermes inference provider's credentials expired or were revoked. A tiered runbook: status and refresh first, a browser account-chooser click only if one is already showing, otherwise hand off to the human.
Use for rare, operator-approved host surgery on a Hermes machine (config, plugins, cron, services) when the HTTP API is proven insufficient. Research first, back up, act minimally.
Use when updating Hermes Agent on one or many hosts. Pre-checks, canary one host then roll, verify versions, and know the rollback path. Never update the whole fleet at once.
Use when wiring secrets for a Hermes fleet: a secrets manager (Bitwarden Secrets Manager, 1Password, command helper, or env), per-host and per-profile credentials, and verifying keys without ever printing them.
Use when wiring or checking a shared external memory backend for a Hermes fleet (Hindsight, Honcho, Mem0 and others), and for recall/retain health checks.
Use after a Grok Bot computer update or reset, or when every Hermes host suddenly fails health with timeouts - restores the private-network path from a recipe kept in persistent storage.