AI coding skills for building agent services on the CEF AI stack. Install into Claude Code, Cursor, Codex, or any AI coding tool that supports the Agent Skills standard.
npx skills add cef-ai/skills| Skill | Description |
|---|---|
| clientsdk | Connect external code to CEF via @cef-ai/client-sdk: send events, query cubbies, subscribe to streams |
| cli | Config schema, deploy/delete/clone commands, local dev server (cef dev), local automated testing, playground testing, environment variables, naming conventions |
| coding | Handler signature, Context API (context.models/.agents/.cubbies/.streams/.rafts/.image/.emit/.fetch/.workspace/.log), entity hierarchy, V8 constraints, orchestration patterns, topology generation |
| inference | context.models.<alias>.infer/.stream and the 16-model catalog (yolo, yoloXL, whisper, whisperTiny, whisperLarge, llm, mistral7b, mistralSmall, qwenCoder, qwenVision, llamaVision, embedding, emotionClassifier, sentimentAnalysis, plateDetector, plateOcr) — per-model input/output schemas, streaming usage, and production examples |
| storage | Cubby API (SQLite query/exec, migrations, sqlite-vec), state machine, dedup, SQL patterns |
| widgets | Build Cere Sandbox widgets that visualize cubby data in ROB: WidgetSandbox / WidgetRuntime contract, manifest shape, custom vs auto-mount rendering, multi-file ES module limits, local testing, troubleshooting checklist |
Skills are automatically discovered by your AI coding assistant based on context. When you ask it to build a CEF agent, call a model, or manage cubby state, the relevant skill loads and provides CEF-specific guidance, code patterns, and production examples.
All CEF handler code must be fully inline (no import or require). The runtime uses V8 isolates. Every skill enforces this rule.