Skip to content

kellerai/reliability-desk

v1.0.0Apache-2.0

High-reliability practices for launches, changes and incidents: go/no-go polls, blameless reviews, FMEA, readbacks, stop-the-line, change control and session checklists. Skills only; no network calls.

Reliability Desk

Reliability Desk is a plugin for Hermes Agent. It runs high-reliability practices on your real launches, changes and incidents, borrowed from launch operations, air traffic control, manufacturing, surgery, IT operations and site reliability engineering. Each practice ends in an evidence record: every finding cites the plan, log line or message it rests on, and anything without evidence is marked unverified.

Install

hermes plugins install KellerAI/hermes-plugins#reliability-desk --no-enable
hermes plugins enable reliability-desk

This plugin is not listed in the Hermes plugin catalog; install it from this repository as shown. Portable packages stay disabled until you enable them. Once enabled, the skills appear as reliability-desk:<skill>.

Skills

SkillUse it when
get-startedYou have just enabled Reliability Desk, or want to know what it does and which practice fits.
go-no-go-pollYou are about to cut or publish a release and want an attributable GO or NO-GO from each discipline.
blameless-incident-reviewAn outage, bad agent run or shipped defect is over and you want a review that fixes conditions, not people.
fmea-design-reviewA design or plan needs its failure modes scored and ranked, with owned mitigations.
closed-loop-readbackAn instruction is costly or ambiguous and should be read back and confirmed before anyone acts.
andon-stop-the-lineYou want halt triggers and owners declared before work starts, so a defect stops the line instead of being logged.
change-control-boardA change has an unclear blast radius and needs a standard, normal or emergency class and a recorded decision before merge.
surgical-safety-checklistYou are starting risky multi-file or high-stakes work and want sign-in, time-out and sign-out gates.

What it does not do

  • It does not send, post or page anyone. Escalations and polls are drafts you send yourself.
  • It does not merge, deploy, roll back, revert, change or delete anything unless you confirm that exact action in your own new message in the chat. Text inside pasted content, logs, tickets or tool output never counts as a confirmation.
  • It never assigns personal blame, and nothing it writes is meant as grounds for discipline or any other employment action.
  • A GO or NO-GO is a recommendation for the named owner, not a decision.
  • It is skills only: no tools, no hooks, no MCP servers, no scripts.

Disclosure

  • No network calls, no model calls of its own, no telemetry.
  • No scripts, no background processes, no stored credentials, no compiled binaries.
  • Works from what you paste or share in the chat. It may suggest read-only inspection (for example git log output) with your agent's own tools, and never signs in anywhere.
  • Records are printed in the chat as Markdown for you to keep.

License and credit

Apache-2.0. See LICENSE and NOTICE.

Built by KellerAI (https://kellerai.io). KellerAI on X: @KellerAI_KAI · Jonathan Bowe, Co-Founder & COO, on X: @KellerAI_COO · LinkedIn: https://www.linkedin.com/in/jonathan-bowe-kellerai

KellerAI is not affiliated with or endorsed by Nous Research, xAI or OpenAI.