aisets

Build, debug, and harden AI task-execution workflows that must call tools accurately across native OpenAI-style tool calling, Codex/Claude Code/Pi-style CLI agents, and local LLMs on mid-range machines. Use when designing or fixing prompts, tool schemas, action-block fallbacks, validation matrices, VLM image grounding, canvas/file/tool orchestration, AI harnesses, or recurring test-improve-learn loops where AI must execute multi-step tasks reliably instead of answering with prose.

Aisets UI design system — token usage, CVA component patterns, Tailwind co-location rules, and pre-delivery checklist. Use this skill whenever editing ANY file under ui/src/, including .tsx, .ts, and style files. Also use when creating new UI components, modifying existing ones, changing colors/spacing/typography, adding animations, or touching anything visual. Even simple one-line UI tweaks should consult this skill — the token and cascade rules catch subtle bugs.

Capture lessons learned from the current session and persist them into CLAUDE.md and .skillshare/skills/design-system/SKILL.md. Use this skill whenever the user says "更新規範", "記取經驗", "update rules", "update SKILL", "寫進 CLAUDE.md", or after a debugging session where new project conventions were discovered. Also use proactively when a bug fix reveals a pattern that should become a rule.