Skills
effect-uai ships a bundle of agent skills
for AI coding agents (Claude Code, Cursor, Continue, …). One main
skill covers the philosophy, primitives, and provider wiring; fourteen
recipe sub-skills each carry one loop-body pattern. The agent only
loads a skill when its description matches your prompt.
Install
Via the skills.sh CLI:
npx skills add betalyra/effect-uaiCherry-pick a single skill instead:
npx skills add betalyra/effect-uai/tree/main/skills/effect-uai-model-retryWhat loads when
The main skill (effect-uai) loads on broad signals like “I’m
building an AI agent in Effect”. Recipe sub-skills load on specific
scenarios:
| When you ask for… | Skill |
|---|---|
| An agent that can call tools | effect-uai-basic-usage |
| Typed JSON output / fill a form | effect-uai-structured-output |
| Stream typed objects as the model writes them | effect-uai-streaming-structured-output |
| Human approval for sensitive tool calls | effect-uai-tool-call-approval |
| Show progress while a tool runs | effect-uai-streaming-tool-output |
| Long-lived chat agent with a queue | effect-uai-agentic-loop |
| Retry rate limits / 5xx with exponential backoff | effect-uai-model-retry |
| Fall back from one provider to another | effect-uai-multi-model-fallback |
| Summarize history when it gets too long | effect-uai-auto-compaction |
| Pause the loop between turns and resume later | effect-uai-pause-resume |
| Stop button / abort the current response | effect-uai-mid-stream-abort |
| Compare answers from multiple models side-by-side | effect-uai-multi-model-compare |
| Models judge each other and pick a winner | effect-uai-model-council |
| Stream the output as SSE / JSONL | effect-uai-modify-output-stream |
Source and per-skill bodies live in
skills/.