Hermes skill · memetic emergence

Hyperlex

Catch language while it is still becoming culture — slang radar with settled forecasts only.

v0.4.2 · offline-first

Hyperlex — memetic emergence

What it is

Hyperlex watches how new phrases move through chats, forums, and meme stacks — what showed up, which family it belongs to, how hard it spreads, and whether it is starting to stick.

Offline-friendly by default, receipt-backed when you care. Brier scores only after a human settles an outcome — no fake accuracy.

In short

Week-one path is a short guided wizard; then demos and scans without API keys on the mock route.

Built with

  • Hermesmain agent that runs the skill
  • Claude Codeoptional second agent host
  • OpenClawagent host listed in the install manifest
  • Pythonlanguage the skill and command line use
  • SQLitedefault local store for searchable phrase history
  • ChromaDBoptional search index, local or cloud
  • Crawl4AIcollects public web pages when online
  • MkDocs and GitHub Pagespublishes docs and a cleaned run history

Changelog

Milestones in plain English, newest first, to Nov 7, 2026

  • Added an optional check that keeps near-duplicate examples out of separate training and test sets.
  • Adopted a clearer set of labels for human rulings, with correction-only replies and no automatic term rejection.
  • Added six new slang families to the family registry and retired the sports-betting family.

  • v0.4.2: Claude Code can now install and run Hyperlex alongside Hermes.
  • Installing, settling forecasts and cloud uploads now stop safely on errors instead of carrying on.
  • The installer can now recover from an interrupted install.
  • v0.4.1: shortened the skill description so Hermes shows the full trigger text.

  • Added a short guided wizard that walks new users through an offline first week.
  • v0.4.0: one command now runs the whole path from collecting text to finished results.
  • Forecasts get accuracy (Brier) scores only after a person settles the outcome, with a tamper-evident receipt log.
  • Launched a public docs site with cleaned run history and an interactive map of slang families.

Roadmap

What comes next. Plans, not promises, and no dates.

  • PlannedReach a stable 1.0 with a fixed skill contract and a long-term analysis archive.
  • In progressOffer an optional setting that copies new local results to the cloud search index after each run.
  • PlannedAdd a faster search option if the phrase collection grows too large for simple lookups.