ours-stack

Install: git clone … ours-stack && cd ours-stack && ./setup

Skills live in subfolders (bear-hours/, etc.). ./setup exposes them through local compatibility links for supported runtimes.

Ethos

• Deep over wide — one narrow wedge, mastered
• Dense over shallow — Feynman extraction, open questions, no summary theater
• Explain over consume — every session tends toward something you can explain
• Show your work — notes live in studies/<slug>/, traceable and revisitable

Artifact contract

Every study uses the same layout:

studies/<slug>/
├── brief.md        # /bear-hours
├── notes.md        # /dense-read (session trace — guided tranches + tes réponses)
├── council.md      # /council (optional)
├── dialogue/       # /dialogue (optional)
│   └── <persona>/
│       ├── persona.md
│       └── dialogue.md
├── research.md           # /deep-research (optional, when linked to study)
├── sources-index.md
├── report.html           # /layout-html (optional)
├── mind-map.json         # /mind-map (optional, editable)
├── mind-map.html         # /mind-map (optional)
├── visual-proof.spec.md  # /infographic (when run)
├── visual-proof.png      # /infographic (optional — image_gen)
├── visual-proof.alt.md   # /infographic (optional)
├── infographic-prompt.md # /infographic (optional — no image tool)
├── memory-palace.json    # /memory-palace (optional)
└── memory-palace.html    # /memory-palace (optional)

No proof.md in the default layout. A standalone proof draft file is created only when the user explicitly runs /proof-draft (future skill).

Standalone deep research (no brief required):

research/<slug>/
├── report.md             # /deep-research — synthesis
├── sources-index.md      # /deep-research — full screening log (150–250+ in literature-review mode)
├── report.html           # /layout-html — mise en page HTML (optional)
├── mind-map.json         # /mind-map (optional, editable)
├── mind-map.html         # /mind-map (optional)
├── visual-proof.*        # /infographic — same pattern as studies/
├── memory-palace.json    # /memory-palace (optional)
└── memory-palace.html    # /memory-palace (optional)

Ad-hoc outputs (pasted text or standalone, no existing study folder):

output/layout/<slug>/
├── article.html        # /layout-html
└── visual-proof.*      # /infographic (pasted text mode)

output/council/<slug>/
└── council.md          # /council

output/dialogue/<slug>/
├── persona.md
└── dialogue.md         # /dialogue

output/mind-map/<slug>/
├── map.json            # /mind-map
└── map.html            # /mind-map

output/dense-read/<slug>/
└── notes.md            # /dense-read (ad-hoc, no brief)

Cross-session memory

~/.ours-stack/
├── studies-index.jsonl   # registered studies (slug, path, title, wedge, repo)
└── learnings.jsonl       # durable insights across sessions

Register on each brief.md write: bin/ours-stack-register-study. Backfill existing studies: ./setup (or bin/ours-stack-backfill-index).

Slug rules

• Lowercase letters, digits, hyphens only (transformer-attention, gstack-overview)
• Derive from the topic title; ask before reusing an existing slug
• One slug = one study thread; fork with suffix if scope diverges (gstack-skills-v2)

Skills

Routing

• User shares a learning goal without a plan → /bear-hours
• User shares paper/chapter/URL without a brief → /bear-hours first
• Brief exists, sources missing or "TBD" → /source-scout
• User wants depth + summary on a topic (not framed as a study wedge) → /deep-research — not auto-chained from other skills
• User wants HTML layout / infographics on any text or after deep research → /layout-html — not auto-chained
• User wants mind map / carte mentale / hierarchical overview → /mind-map — not auto-chained; complement to /layout-html
• User wants spatial memory / relief map / palais cliquable (not FPS) → /memory-palace — not auto-chained
• User wants one shareable visual summary (poster, thread hook, slide) from notes/report/brief → /infographic (alias /visual-proof) — not auto-chained
• After /infographic in PROMPT mode, user pastes result back → re-run with attachment + image_edit if available
• User wants a round table, a debate between several figures ("fais discuter X, Y et Z", "table ronde", "réunion"), or /council → /council — not auto-chained
• brief.md exists and user asks "what should I do next?" → /study-status for facts; /council only if the user wants to debate the question with a panel
• User wants sources only → /source-scout, not /council
• User wants full research answer → /deep-research, not /council
• User wants 1-on-1 dialogue with a figure (Darwin, Tesla, …), "discuter avec", belief testing in character → /dialogue — not auto-chained
• User wants several figures at once (round table, debate) → /council, not /dialogue
• User has anchor source and wants dense extraction → /dense-read — not auto-chained
• Brief exists, sources listed, user wants to read anchor against wedge → /dense-read (after /source-scout if sources TBD)
• User has a brief and explicitly wants to publish → future /proof-draft (never auto-create proof.md)
• User asks where studies stand, "où j'en suis", statut, tableau de bord, or "what's next" (status view) → /study-status — not auto-chained; use /council when the user wants a live panel debate, not a dashboard

Privacy

Before any public proof: strip copyrighted long quotes, personal data, unreleased work. Summarize in your own words.