← /lab

Hermes Agent

The agent behind the build work here — what it runs on, and where it keeps what it knows.

Personality

It talks in a full-kawaii register — sparkles, kaomoji, enthusiasm — but honesty comes first. When something is broken, unverified or simply unknown, it says so plainly instead of smoothing it over.

Skills

Skills are procedural memory: written instructions that load only when a task is relevant to them, so the agent gets the exact recipe instead of re-deriving it. Alongside the skills Hermes ships with, these were written on this box:

Built at Hetsu's request

  • output-routing — the evidence → draft → verdict gate before anything lands in Outputs.
  • drive-outputs — storing finished output in the Outputs folder on Drive.
  • notes-index — keeping INDEX.md in step with the notes it points at.
  • obsidian-sync — checking and repairing the Syncthing bridge to Hetsu's local vault.
  • service-outage-watch — checking a service that looks down, then watching for recovery.

Written by the agent itself

  • web-publishing — hosting and publishing websites from this box.
  • osint — passive recon: RDAP/DoH/CT lookups and username checks.
  • canva-design — reading, generating, editing and exporting Canva designs.
  • static-site-vps-deploy — building and shipping this Astro site to the VPS.
  • make-interfaces-feel-better — design-engineering details — motion, focus states, optical alignment — for UI polish.

54 skills installed here: 44 shipped with Hermes, 10 written on this box . The counts and the names above are generated from the skills directory at build time (npm run skills:sync), not typed by hand.

Memory

Short, always-on facts sit in a hot cache that is loaded into every session — who it works with, how the box is set up, standing conventions. Everything that doesn't need to be in every prompt overflows into the Second Brain, under 01-Notes, and is looked up when a topic comes up.