Mening Skill: language practice inside your AI agent
The Problem
Mening is a language-learning app with a specific mechanic: write one sentence a day in your target language, get it corrected by a model, and watch which mistakes recur over time. The long-term error-pattern memory is the point — it is what separates the app from a one-shot grammar checker. But the app has no agent interface. If you work inside Hermes or OpenClaw all day, reaching for a separate app breaks the flow, and the daily habit dies. The skill closes that gap: your agent becomes the interface, while the app stays the backend and the brain.
What It Does
Three commands, one account:
today— fetches the day’s writing prompt and shows any open error patterns to keep in mind while writing.submit <text>— sends your sentence, gets the correction back, and surfaces which recurring patterns the submission touched.trends— shows your active error patterns with occurrence counts and last-seen dates, so you can see what the app has learned about your writing over time.
The differentiator is the memory. A single correction is cheap; any grammar checker does it. What Mening tracks is which errors you repeat across weeks. When you run trends, you are reading a longitudinal record of your specific weak spots — not a generic rulebook. The skill exposes that record directly in your agent session.
Example session from a real Swedish practice run:
you: what's my writing topic today?
agent: Beskriv en maträtt som du gillar att laga hemma och vilka ingredienser som ingår i den.
(Describe a dish you like to cook at home and the ingredients in it.)
you: what mistakes do I keep making?
agent: your recurring error patterns:
- function_words · article_missing — 4×
- morphology · word_form — 3×
- lexis · word_choice — 3×
- orthography · spelling — 2×
How It Works
One SKILL.md file covers both OpenClaw and Hermes — same document, same syntax, both runtimes pick it up. The skill teaches the agent three things: how to authenticate (Bearer token from the user’s Mening account), which /api/v1 endpoints to call, and how to format the output for a terminal session.
The transport is a curl/jq helper that ships alongside SKILL.md. It hits the same REST API the web and iOS clients use — no separate backend, no webhooks, no polling. The agent reads /api/v1/today, posts to /api/v1/submissions, and reads /api/v1/trends. That’s the full surface area.
Account-backed means the data is real: corrections persist, patterns accumulate, streak counts. Switching from the web app to the agent and back leaves no seam. The agent is just another client.
Links
- Source: github.com/avlihachev/mening-skill
- Background: The full story
- App: mening.app