ours-stack

Agent skills for autodidacts — deep learning, dense papers, explain what you learn.

By School of the Bear.

Inspired by gstack, for learning instead of shipping code.

Install

git clone https://github.com/loursCreatif/ours-stack.git ours-stack
cd ours-stack
./setup

Run ./setup once — it creates the local compatibility links used by supported skill runtimes.

Workspace: open this repo (or any repo with studies/), not /. If /source-scout cannot see local files, run ./setup and restart the session.

Use

/bear-hours         → frame the topic (5 questions) → studies/<slug>/brief.md
/source-scout       → 3 sources mixtes     → article + YouTube + joker dans brief.md, ouvertes dans le navigateur
/dense-read         → read anchor source   → studies/<slug>/notes.md
/deep-research      → research + summary   → research/<slug>/report.md (standalone)
/layout-html        → page HTML autonome   → report.html / article.html (typo + SVG fidèles)
/mind-map           → carte mentale        → mind-map.json + mind-map.html
/memory-palace      → carte relief oblique  → studies/<slug>/memory-palace.html
/infographic        → visuel mémorable     → visual-proof.png or infographic-prompt.md
/council            → table ronde vivante  → studies/<slug>/council.md
/dialogue → talk with a figure  → studies/<slug>/dialogue/<persona>/
/study-status      → study dashboard      → chat only (read-only)

/deep-research is standalone — invoke it directly on any question; no pipeline required. /layout-html turns finished text into one self-contained HTML page — editorial typography, SVG figures built only from source data; opens offline (file://). No new research. /infographic (alias /visual-proof) turns study notes or reports into one poster-style visual — generates PNG when an image model is connected, otherwise exports a paste-ready prompt for Midjourney, DALL·E, or Flux. /dialogue is standalone — 1-on-1 Socratic dialogue; use /council for a live round table with several figures (the user participates). Alias: /dialogue-historique. /study-status is read-only — criteria, artifacts, and suggested next step per study; no file output.

Cross-project memory lives in ~/.ours-stack/studies-index.jsonl. ./setup backfills existing studies/ into the index.

Examples:

/bear-hours I want to understand how gstack structures its skills
/source-scout robotique-assemblage-structurel
/dense-read biomimetisme-locomotion-chantier
/deep-research What are the best approaches for structural assembly robots on construction sites?
/layout-html robots-assemblage-structurel-chantier
/layout-html studies/biomimetisme-locomotion-chantier/notes.md
/mind-map biomimetisme-locomotion-chantier
/mind-map regenerate studies/biomimetisme-locomotion-chantier/mind-map.json
/memory-palace biomimetisme-locomotion-chantier
/memory-palace research/robots-assemblage-structurel-chantier theme construction
/infographic biomimetisme-locomotion-chantier
/council fais débattre Tesla, Edison et Elinor Ostrom sur l'énergie décentralisée
/council robotique-assemblage-structurel
/dialogue discuter de l'évolution avec Darwin
/dialogue robotique-assemblage-structurel
/study-status
/study-status robotique-assemblage-structurel

Test v1: open studies/robotique-assemblage-structurel/brief.md — ## Source material should list an anchor + core sources after scout.

Details: AGENTS.md

Develop

Clone anywhere, run ./setup — it creates the runtime compatibility symlinks. Edit a */SKILL.md, test in your agent, commit.

Add a skill: copy a folder, write SKILL.md, add one line to AGENTS.md and this README.

Ethos

• Deep over wide — one wedge at a time
• Dense over shallow — explain it simply or you don't know it yet
• Explain over consume — every session tends toward something you can explain
• Show your work — everything lives in studies/

License

MIT — LICENSE