Archetype AI Agent Skills
Agent skills for building applications with Archetype AI's Newton — a real-time sensor intelligence platform that understands physical world data through foundation models.
Inspired by mongodb/agent-skills.
Skills
| Skill | Description |
|---|---|
| atai-newton-fusion-model | Call the Newton C 2.6 fusion model on /query with text, image, or video in one stateless POST — the first C checkpoint to reason over video frames via /query |
| atai-newton-omega-model | Get time-series embeddings from the Omega encoder (OmegaEncoder::omega_embeddings_1_4) over /query — one stateless call per channel, fanned out in parallel — for client-side KNN classification, anomaly scoring, and similarity search |
| atai-newton-omega-model-data-prep | Clean, split, and featurize multivariate time-series data before the Omega model — gap-aware blocking + imputation, out-of-time train/test split, and joint-state (X, y) featurization |
| atai-design-system | Build a Newton demo front-end with the Archetype AI Design System — scaffold via the ds CLI (@archetypeai/ds-cli) and compose the published Svelte 5 primitives + OKLCH tokens (@archetypeai/ds-{lib-tokens,ui-svelte-console,ui-svelte-labs}) instead of hand-rolling UI |
More skills are in review and will be added to this table as they land.
Building with the Design System
When starting a new demo front-end, don't hand-roll the UI — scaffold with the atai-design-system skill and build on top of what it generates. The ds CLI stands up a SvelteKit + Tailwind v4 project wired to the published Archetype AI Design System packages (branded tokens, fonts, and Svelte 5 primitives), and writes an agent config (CLAUDE.md/AGENTS.md + ds-manifest.json) that documents every available component, its usage recipes, and variant axes.
Use that scaffolded project as the baseline for your demo: the tokens, brand styling, and component library are already in place, so you can focus on the Newton integration (fusion or Omega calls via the model skills) and the demo-specific views.
Example apps
End-to-end demos built on these skills, on the Direct Query API (SvelteKit unless noted):
| Demo | Skills | What it demonstrates |
|---|---|---|
| Traffic Monitor | atai-newton-fusion-model |
Live Caltrans CCTV feed → C 2.6 vision. Samples a short burst of frames per interval and sends it as one multi-frame /query clip to reason over traffic flow, incidents, and conditions. |
| Wildfire Watch | atai-newton-fusion-model |
1,200+ ALERTCalifornia cameras → C 2.6 vision. Per-camera smoke/fire/haze detection, a zone scan that batches frames per /query, and text-only zone Q&A. |
| Earthquake monitor | atai-newton-fusion-model |
Live USGS feed → C 2.6 text reasoning. Formats the current quakes as structured text and asks Newton (stateless /query) to surface aftershock sequences, spatial clustering, and ranked regional risk — alongside an interactive world map. |
| SWaT water treatment | atai-newton-omega-model + atai-newton-fusion-model |
Six-stage plant anomaly detection: Omega per-channel embeddings + client-side KNN, plus C 2.6 text reasoning for operator action suggestions. |
| Wind turbine monitor (Python/Flask) | atai-newton-omega-model + atai-newton-omega-model-data-prep |
Penmanshiel wind-farm SCADA anomaly detection: Omega per-channel embeddings + local KNN against a leakage-free n-shot library precomputed offline. Replays 3 months of telemetry and detects a real frequency-converter fault on one turbine against its healthy peer. |
| Drilling state monitor | atai-newton-omega-model + atai-newton-omega-model-data-prep |
Equinor Volve North Sea well SCADA: per-channel Omega embeddings + local KNN classify each window as drilling / not-drilling, against a leakage-free n-shot library precomputed offline from held-out reference wells. Replays real well telemetry with live accuracy vs ACTC ground truth. |
| Grid monitor | atai-newton-fusion-model |
Live CAISO power-grid feed → C 2.6 text reasoning. Formats 5-minute demand/supply data as structured text and asks Newton (stateless /query) about duck-curve dynamics, evening ramp, renewable share, and grid-stress risk, with supply/demand charts. |
| WiFi occupancy monitor | atai-newton-fusion-model |
Residential gateway WiFi telemetry → C 2.6 text reasoning. Sends an anonymized 15-minute per-device flow/byte/protocol snapshot as JSON via stateless /query; Newton infers home occupancy (OCCUPIED … EMPTY) from traffic patterns alone — no device-type labels, no online flag. |
Quick Start
Any coding agent (recommended)
Install via the skills.sh CLI — it detects your coding agents (Claude Code, Cursor, Codex, Copilot, and 20+ others) and installs to the right location for each:
# Install into the current project
npx skills add archetypeai/agent-skills
# Or install globally (available in all projects)
npx skills add archetypeai/agent-skills -g
Manual (Claude Code)
# Add as global skills (available in all projects)
cp -r skills/* ~/.claude/skills/
# Or add to a specific project
cp -r skills/* your-project/.claude/skills/
Invoke a Skill
/atai-newton-fusion-model # Multimodal (text/image/video) queries on the C 2.6 fusion model
/atai-newton-omega-model # Omega time-series embeddings + client-side KNN via /query
/atai-newton-omega-model-data-prep # Clean / split / featurize time-series before the Omega model
/atai-design-system # Scaffold + build a Newton demo front-end with the Archetype AI Design System
Architecture
The skills here target Newton's Direct Query API — one stateless POST to /query per request, no session lifecycle, no batch jobs, no SSE plumbing:
Text: POST /query ──────────────────────────────────► response
Image(s): Upload file(s) → POST /query (file_ids) ───────► response
Video (.mp4): Upload file → POST /query + max_frames ────────► response
Video frames: POST /query (frames + query_metadata) ─────────► response
The reference scripts are built on the official Archetype AI python client (pip install archetypeai).
API Base URL
https://api.u1.archetypeai.app/v0.5
License
Apache-2.0
No comments yet
Be the first to share your take.