A persistent learning workspace. One directory per topic, taught across many sessions.
Learning a topic properly takes more than one sitting, and the thing that usually gets lost between sittings is the state: what was already covered, what stuck, what turned out to be a misconception, which sources were worth trusting. This repository holds that state.
Topics live under learning/, grouped by domain. Each is a track with a stated mission, an ordered set of short lessons, cheat sheets built for lookup, and a record of what was demonstrably learned. The teach skill reads the track at the start of a session, writes one lesson, and records only what the session earned.
- An agent that can read
AGENTS.md. Theteachskill it needs ships with the repository, at.agents/skills/teach/.
cp -r templates/learning-workspace learning/<domain>/<topic-slug>Then run the teach skill and name the topic. It interviews you for the mission, writes README.md, and produces the first lesson.
Run the teach skill again. It picks up from the track state, so there is no need to recap what you covered last time.
.
├── learning/
│ ├── README.md # index of topic tracks
│ └── <domain>/<topic>/ # one track per topic
├── templates/
│ └── learning-workspace/ # copy source for a new topic
└── .agents/
└── skills/teach/ # the format spec every track follows
Conventions for agent behavior live in AGENTS.md.
See the org-wide CONTRIBUTING.md.
Two licenses, because the repository holds two different things.
- Learning content (everything under
learning/andtemplates/): CC BY 4.0. Copy it, translate it, teach from it, build on it, commercially or not. Just say where it came from. - Software and configuration (build config, workflows, stylesheets): Apache-2.0.
One exception: everything under
.agents/skills/teach/ is MIT, derived from
mattpocock/skills via
PromptPasture/agent.md. Its
attribution lives in the front matter of SKILL.md, which the other files in
that directory were split out of.