Documentation Index
Fetch the complete documentation index at: https://docs.archetypeai.app/llms.txt
Use this file to discover all available pages before exploring further.
What are Agent Skills?
Agent Skills are packaged, reusable instructions that teach an AI coding agent how to handle a specific Newton task — for example, “create a batch classification job and poll until it completes” or “upload a 1 GB file via multipart presigned URLs.” Each skill is a folder containing a SKILL.md (plus optional supporting files) that the agent loads on demand.
Skills give agents the institutional knowledge you’d otherwise have to repeat in every prompt: which endpoint to call, what the schema looks like, the empirical gotchas, the patterns we’ve already proven on real datasets. They’re the difference between “explain the batch API” (you do the synthesis) and “run a Machine State batch job on this CSV” (the agent already knows how).
The canonical source is the archetypeai/archetypeai-agent-skills repository.
Install
Claude Code
Copy the skills folder into Claude Code’s skills directory:
# Clone the repo
git clone https://github.com/archetypeai/archetypeai-agent-skills.git
cd archetypeai-agent-skills
# Install globally (available in every project)
cp -r skills/* ~/.claude/skills/
# Or install into a single project
cp -r skills/* /path/to/your-project/.claude/skills/
/newton-setup
/newton-machine-state-batch
/newton-activity-detection-batch
Other agents
The SKILL.md format is portable. The same files work with any agent runtime that loads skills from a directory — Cursor, Codex, Gemini CLI, Cline, and the broader skills ecosystem. See each agent’s docs for the directory it reads from.
Available skills
Getting started
| Skill | When to use |
|---|
newton-setup | Configure Newton API access, environment variables, and SDK initialization. Use first when bootstrapping a new project, or when an existing project hits auth errors. |
newton-models | Prod-scoped catalog of every Newton C / Omega / vision-lens model identifier currently exposed on api.u1.archetypeai.app, the API surfaces each is reachable from (/query, batch pipelines, Lens sessions), and recommended defaults per use case. Use when picking a model for a new project, when an identifier returns invalid_model_version, or as the source-of-truth registry for which skills to update when a new checkpoint ships. |
Real-time inference (Lens sessions, /query, streaming)
| Skill | When to use |
|---|
newton-machine-state | Classify time-series sensor data with n-shot examples via the Machine State Lens. Covers session lifecycle, focus CSV uploads, sliding-window config, and SSE result parsing. |
newton-machine-state-direct-query | Stateless per-window n-shot classification by calling /query directly with the Omega encoder, then running KNN locally — no Lens session, no SSE, no warmup. Use when you need direct access to the embeddings (for visualization, custom distance metrics, or per-request handlers) or want a simpler failure model than streaming Lens. |
newton-activity-monitor | Vision-based analysis and natural-language Q&A using the Activity Monitor Lens — charts, dashboards, camera feeds, screenshots. Two lenses are mounted on prod, one pinned to Newton C 2.5 (recommended) and one to C 2.4 (legacy). Covers screenshot-to-video pipeline, focus text, and SSE handling. |
newton-query-prompting | Prompt-engineering patterns for the /query endpoint across text, structured-data (CSV / JSON), and image inputs — structured-output enforcement, contamination avoidance, constraint tables, topology validation, and the gotchas around which input modes inject contents into the prompt (and which don’t). |
newton-sensor-streaming | Real-time sensor data ingestion — BLE heart rate monitors, OBD2 adapters, serial devices, IoT sensors. Covers Web Bluetooth, Web Serial, OBD2 protocols, and the NewtonStreamManager singleton pattern. |
Batch processing (asynchronous jobs)
| Skill | When to use |
|---|
newton-batch-upload | Upload files larger than 255 MB via multipart presigned URLs (/files/uploads/initiate → PUT → complete). |
omega-preflight | Predict whether a binary time-series dataset will clear ~70% accuracy on omega_1_4_base + the machine-state-classification pipeline before committing to a full batch run. Runs 10 fast local static checks (schema, timestamp monotonicity, missing values, constant columns, scale heterogeneity, n-shot support, class balance, window-vs-sampling translation, accuracy prior) and an optional --pilot mode that hits the real batch API on a held-out slice. v1 is binary-only. |
newton-machine-state-batch | Run the machine-state-classification pipeline as an async batch job — millions of rows of sensor data, n-shot KNN, omega_1_4_base by default. Covers config optimization and the within-distribution vs cross-condition accuracy pitfall. |
newton-activity-detection-batch | Run async text/vision generation on the Newton C model via the activity-detection pipeline. Covers InferenceRecord schema (text / image / video), the ~4K-token quality cliff for CSV-heavy inputs, MapReduce / hierarchical-reduce patterns, and the silent join bugs to watch for. |
Local & offline
| Skill | When to use |
|---|
omega-local | Generate time-series embeddings locally from a vendored Omega 1.3 checkpoint — no cloud roundtrip. Useful for custom downstream models (KNN, Isolation Forest, t-SNE/UMAP) or for experimenting with windowing/normalization strategies the cloud API doesn’t expose. |
Example projects
These public repos demonstrate the patterns covered by the skills above — clone them as starting templates:
- archetypeai-batch-examples-volve — batch upload, inference, and evaluation with Volve drilling data (Machine State + Activity Detection).
- archetypeai-batch-examples-ghost-iot — 1 GB WiFi flow CSV folded into 9 daily narratives via six MapReduce stages on the
activity-detection batch pipeline. Reference implementation for newton-activity-detection-batch — cliff sweep, hierarchical reduce, N-way positional split, content-key joins.
- archetypeai-swat-demo — 6-stage water-treatment-plant anomaly dashboard with parallel per-stage Machine State Lens sessions and
/query-generated operator suggestions. Reference for newton-machine-state and newton-query-prompting.
- archetypeai-swat-demo-direct-query — same SWaT 6-stage dashboard but rebuilt on the stateless
/query + local KNN pattern. PCA-2 + UMAP-2 visualization of the live embeddings, leave-one-out KNN accuracy badges per stage, and head-to-head model-comparison scripts. Reference for newton-machine-state-direct-query (and a worked example for newton-models’s pre-promotion verification flow).
- archetypeai-wildfire-demo — wildfire detection across 1,200+ ALERTCalifornia cameras with Newton vision. Reference for
newton-activity-monitor.
- archetypeai-earthquake-demo — real-time USGS earthquake analysis + Newton text reasoning.
- archetypeai-grid-demo — California power grid monitoring (CAISO supply/demand) + Newton text reasoning.
- archetypeai-drilling-demo — drilling state classification from 14 North Sea wells + Newton Machine State Lens (SSE streaming).
If you build something with these skills, open a PR against archetypeai-agent-skills to share your patterns — the empirical gotchas in each SKILL.md came from real production runs and we’d love more. 
