agentleFS
loursCreatifloursCreatif/ours-stack/memory-palace/SKILL.md

memory-palace

Skill1 starsChanged 4 months agoMITView on GitHub

Carte en relief oblique (style Google Maps 3D) — cliquer bâtiments et zones pour ancrer concepts (method of loci). Drill-down site → intérieur, panneau latéral. Pas de déplacement FPS. Reads brief.md, notes.md, research.md, report.md. Outputs memory-palace.json + HTML (Three.js vendored). Triggers: "memory palace", "palais de mémoire", "relief map", "carte oblique", /memory-palace. No new research.

4 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

10 rules. 0 are common, 2 are distinctive, and 8 are very specific to this project.

  • 0

    Common

    Most of the 3 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 8

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Use AskUserQuestion — one call, wait for answer before continuing.

    line 53
  2. Distinctive:

    STOP: after the call; wait for answer

    line 59
  3. Very specific:

    If custom selected: wait for user's next message.

    line 74
  4. Very specific:

    Default when skipped: corridor.

    line 76
  5. Very specific:

    Guard: if no resolvable text and no pasted concepts → ask for slug, path, or concept list.

    line 95
  6. Very specific:

    Pre-flight: if only brief.md with no notes → warn palace will be thin.

    line 103
  7. Distinctive:

    Keep all (warn larger file)

    line 111
  8. Very specific:

    Follow references/palace-schema.json (view_mode: relief).

    line 127
  9. Very specific:

    If fetch fails (no network), fall back to iso-SVG HTML (--2d-only) and tell user to run fetch later for relief 3D.

    line 145
  10. Very specific:

    Add room objects from loci[].objects (simple BoxGeometry props)

    line 159

“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Memory Palace

Transform study concepts into an oblique relief map (method of loci) — click buildings, drill into interiors, read concepts in a side panel. No FPS movement. No research.

When to use (vs other skills)

SkillResponsibilityDoes NOT
/mind-map2D tree, editable JSONSpatial relief map
/layout-htmlLinear article + SVGSpatial map
/memory-palaceThis skill — relief map + drill-down lociNew sources, WASD navigation

Optional pipeline (manual only — never auto-chain):

/bear-hours → [dense read] → /memory-palace
/deep-research → /memory-palace

Hard rules

  1. No research — never WebSearch / WebFetch. Concepts must trace to input files.
  2. Wedge lock — if brief.md exists, exclude out-of-wedge concepts from default path[].
  3. Provenance — every concept: source.file, source.section, optional anchor.
  4. Cap — default 8–16 loci; AskUserQuestion before exceeding 20.
  5. Two artifacts — memory-palace.json (manifest) + memory-palace.html (experience).
  6. Self-contained HTML — Three.js from references/vendor/three.min.js, inlined; no CDN.
  7. Iso-SVG fallback — isometric click map (same drill-down) when no WebGL, mobile, or prefers-reduced-motion.
  8. Relief UX — oblique map by default; optional guided ground tour on rails (Visite guidée); pan/zoom on site level only; click building → interior; no WASD / pointer-lock / free movement.
  9. Language — match source (lang on <html>).
  10. Open browser — open <path> on macOS after write.
  11. Theme first — AskUserQuestion before reading content (unless smart-skip).

Step 0: Theme — AskUserQuestion (always first)

Use AskUserQuestion — one call, wait for answer before continuing.

Title: Memory Palace — Theme

Hard rules:

  • Never list theme options as plain chat bullets — always AskUserQuestion
  • STOP after the call; wait for answer
  • Smart-skip — if user already named a theme (museum, chantier, corridor, library, or ≥5 words describing a place), skip this step

Question: "Quel décor pour ancrer tes concepts ? (method of loci — un parcours fixe aide la mémoire)"

Options (exactly 5):

idlabel
corridorCorridor classique — pièces en ligne, portes numérotées (défaut)
museumMusée — salles thématiques, vitrines = sous-concepts
constructionChantier / atelier — zones, équipements (robotique, BTP)
libraryBibliothèque — rayons = thèmes, livres = citations
customCustom — je décris le lieu dans le prochain message

If custom selected: wait for user's next message. Map to nearest archetype topology; store description in manifest theme_notes.

Default when skipped: corridor.

Load references/palace-archetypes.md for topology before Step 3.

Step 1: Intake — detect input mode

Resolve input in this priority order:

ModeTriggerSourceDefault output
A — Study slugslug or studies/<slug>/brief.md → notes.md → research.mdstudies/<slug>/memory-palace.{json,html}
B — Research slugresearch/<slug>/report.md (+ brief.md via Study link)research/<slug>/memory-palace.{json,html}
C — File pathpath to .mdthat file<dir>/memory-palace.{json,html}
D — Pasted conceptsuser lists concepts in chatmessage bodyoutput/layout/<slug>/memory-palace.{json,html}

Slug derivation (mode D): lowercase-hyphen from title or first heading; ask once if ambiguous.

Custom output: user may specify → out.html — write HTML there; JSON alongside with same basename unless user says otherwise.

Guard: if no resolvable text and no pasted concepts → ask for slug, path, or concept list. Do not suggest /deep-research unless user wanted more content first.

Step 2: Extract concepts

Load references/concept-extraction.md. Read source files. Extract only what exists.

Minimum: ≥5 concepts or abort with guidance ("add notes.md or paste concepts").

Pre-flight: if only brief.md with no notes → warn palace will be thin; proceed only if ≥5 extractable concepts.

If >16 concepts extracted — AskUserQuestion:

Title: Memory Palace — Density

Options:

  • Trim to top 12 by importance
  • Keep all (warn larger file)
  • I'll pick concepts in chat

Step 3: Map loci (relief site)

Load references/loci-mapping.md + references/palace-archetypes.md.

  • path[] = ordered building ids on the site map (5–12)
  • Each locus: kind: building, footprint: [x, z, w, d], height, interior.zones, interior.hotspots
  • Each locus: optional scene — vivid mental image (scene.image, 1–2 sentences) + 2–3 sensory hooks (scene.senses: « Son — … », « Odeur — … », « Texture — … ») encoding the concept for method-of-loci recall
  • Entrée = wedge + why now; zones terrain/mécanismes/robots = findings; bureau = beliefs
  • blocked: true buildings = open questions (click → panel only, no interior)
  • Zone footprints relative to building interior (0…building size)

Step 4: Write memory-palace.json

Follow references/palace-schema.json (view_mode: relief). Write to resolved output dir.

Overwrite without asking unless file modified <1h ago and user didn't say overwrite.

Show user a compact summary: concept count, loci count, path order (one line).

Step 5: Compose memory-palace.html

Load references/navigation-ux.md. Compose via script (preferred).

Before composing:

bash memory-palace/scripts/fetch-three.sh

Pinned version: Three.js 0.160.0 (last npm release with build/three.min.js IIFE). Override: THREE_VERSION=0.160.0 bash …/fetch-three.sh.

If fetch fails (no network), fall back to iso-SVG HTML (--2d-only) and tell user to run fetch later for relief 3D.

Compose (agent-authored or helper script):

python3 memory-palace/scripts/compose-html.py studies/<slug>/memory-palace.json
# 2D only: python3 memory-palace/scripts/compose-html.py … --2d-only

Manual path:

  1. Read memory-palace/references/vendor/three.min.js → inline in first <script> block
  2. Embed full JSON as const PALACE_DATA = …
  3. Inline scene bootstrap from html-template.md § Scene JS
  4. Add room objects from loci[].objects (simple BoxGeometry props)
  5. Theme colors from archetype table
  6. HUD, overlays, 2D plan, footer with source files + study path

If user said 2d only: --2d-only or window.FORCE_2D = true; skip Three.js inline (smaller file).

Step 6: Deliver

open "<absolute-path>/memory-palace.html"

Tell the user:

  1. Output paths (.json + .html)
  2. Theme used
  3. Locus count + concept count
  4. Memory walk script — 30-second verbal path through buildings in path[] order
  5. Source files used
  6. Reminder: recall the path, not just the visuals
  7. Optional: /layout-html for 2D report; re-run with overwrite after new notes

Escape hatches

User saysAction
2d onlyIso-SVG map only (no Three.js)
overwriteReplace JSON/HTML without confirmation
slug onlyTry studies/<slug>/ then research/<slug>/
file pathMode C
pasted conceptsMode D — output/layout/<slug>/
theme in messageSmart-skip Step 0
museum / construction / etc.Smart-skip theme question

Self-check before delivery

  • No WebSearch / WebFetch
  • Every concept label grep-able in source files
  • path[] length ≥ 5 and ≤ 20
  • out-of-wedge excluded from path (if brief exists)
  • JSON validates against palace-schema.json structure
  • HTML offline-ready (no CDN links)
  • view_mode: relief in JSON; buildings have footprint + interior
  • Iso-SVG fallback OR Three.js relief inlined
  • No WASD / pointer-lock in composed HTML
  • lang matches source
  • Files written + browser opened

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.