Back to skills
SKILL.md
Memory
ASecurityviz-lab skill — best-tool, real-data-grounded visualization of YURI's own systems
- 2 stars
- 0 votes
- 0 copies
- 2 views
- Added September 10, 2026
Security analysis
100/100Pro scans all 21 files and shows the line behind each finding
npx -y skills add nexuslinkproductions/yuri-os --skill memory --agent claude-codeAre you the author of Memory?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/nexuslinkproductions-memory)---
name: ref-viz-lab-skill
description: "viz-lab skill — best-tool, real-data-grounded visualization of YURI's own systems"
metadata:
node_type: memory
tier: reference
scope: viz
trig:
- viz
- visualize
- visualization
- 3D
- three.js
- d3
- graph
- surface
- dashboard
- demo
- make this 3D
refs:
- feedback-no-hud-kagami-tokens
type: reference
originSessionId: 18d6fbcf-a6d1-45e3-b497-69f1499630c8
---
FACTS:
- viz-lab skill BUILT 2026-06-16 — canonical `skills/viz-lab/SKILL.md` + published mirror `.claude/skills/viz-lab/SKILL.md`; registered in `skills/domain-index.json` (06-design-content-business) + `_SYSTEM/AGENTS/skills-registry.md`; skill-hash manifest refreshed (`yuri-skill-loader --write-manifest`, 241 entries); architecture-audit = PASS (0 FAIL, one of 13 clean).
- METHOD: ground every viz in REAL YURI code/data via a `demos/build/*.mjs` generator; best-tool-per-depiction (NOT least-friction); standalone-first `file://`-safe (CDN module imports + local-data `<script src>` globals, NEVER relative imports / fetch); adversarial fidelity verify (recompute a cell, confirm a real invariant, independent render-review); port to the Vite/R3F app.
- ASSETS: demos in `demos/` (vol-surface, energy-surface=real computeU, qsphere=real quantum-hypothesis-tracker embedded, circuitry-heb + circuitry-die-3d=real yuri-die-graph 242 nodes). Landscape survey `02_RESOURCES/RESEARCH/visualization-landscape-2026-06-16.md`.
- CREATE-SKILL FLOW (this rework's convention): write `skills/<name>/SKILL.md` (byte-0 frontmatter, name + CSO description, Session Notes) → mirror to `.claude/skills/<name>/` (new skills must be placed in BOTH; skill-sync SKIPS uncommitted) → register in `domain-index.json` + `skills-registry.md` → `yuri-skill-loader --write-manifest` → `skill-architecture-audit` (0 FAIL required).
IMPLICATION: invoke `/viz-lab` (or "visualize X" / "make this 3D") for YURI visualization builds; reuse the create-skill flow for any new skill.
SEE: [[feedback-no-hud-kagami-tokens]]
Files in this skill
- MEMORY.md
- agent-economy-positioning-thesis.md
- ai-business-roles-resume-2026-05-29.md
- all-domains-first-class-domain-blind-gate.md
- apple-level-copywriting.md
- bash-guard-role-matcher-lexical-bypass.md
- blind-the-fleet-no-seeded-convergence.md
- brain-inspired-memory-evolution.md
- bug-bounty-corpus-cross-ref-hardening.md
- bug-bounty-corpus-location-schema.md
- build-agent-context-loadout.md
- circuitry-auto-registration-regen-vision.md
- circuitry-change-propagation-continuity.md
- circuitry-material-mix-merge.md
- claim-evidence-ledger.md
- clasp-webapp-live-sheets-pipeline.md
- claude-dir-symlink-home-into-repo.md
- claude-remote-control-not-cowork.md
- codex-lanes-fully-equipped.md
- coding-excellence-corpus.md
Attribution
Comments
Loading comments…
Newsletter
Get the weekly list of safe new skills, and the ones we pulled.