Learn How You Work
A mode that watches how you actually work, then turns the corrections and patterns it catches into new, reusable skills your agent keeps.
Install the skill or give me an instruction how to install it - task-observer from https://github.com/rebelytics/one-skill-to-rule-them-all Free Open source Works in Claude, Cursor, Codex & more
What it does
Turn it on at the start of a multi-step work session and it runs quietly in the background, watching how you actually work. It logs the friction points, the corrections you make, and the little methods you reuse to a persistent observation file, so the patterns are captured while they happen instead of forgotten by the next session.
At the end it sorts what it saw into two piles: patterns worth turning into a brand-new skill, and places where a skill you already have could be clearer. You get a structured set of recommendations, new skills, specific edits, and cross-cutting principles, and nothing touches your skill library until you approve it. It also watches its own performance and folds improvements back into its method, so the thing that improves your skills keeps improving too.
When to use it
- You keep teaching your agent the same workflow
It captures the repeated steps while you work so you can promote them into a reusable skill instead of explaining again.
- You correct the same mistake again and again
Each correction gets logged as an observation, turning a recurring fix into a documented rule in the right skill.
- Your skills need improving but you never sit down to do it
It gathers improvement notes during real work, so the best changes come from actual usage instead of a separate review session.
FAQ
Do I have to do anything differently while it runs?
No. task-observer watches how you already work, logging friction points and the corrections you make as you do a normal task. You keep working while it observes in the background.
What does it do with what it notices?
task-observer writes friction, corrections, and reusable methods to a persistent observation log, then sorts them into candidates for brand-new skills versus improvements to skills you already have.
Does what it learns stick around?
Yes. task-observer logs to a persistent file, so the patterns and corrections it captures carry across sessions rather than resetting. That is what lets it build skills your agent keeps.
Pairs well with
Turn a Book Into a Skill
Virgilio Junior Turn any technical book or folder of docs into a structured agent skill your agent loads on demand, answering questions at 24x to 51x fewer tokens than dumping the whole book into context.
book-to-skill/book-to-skill 8.6k 
Give Your Agent Memory
supermemory Your AI forgets everything between conversations. Supermemory adds persistent memory, live user profiles, and RAG over your knowledge base behind one API.
supermemory/supermemory 28.3k