Agent Skills

This repository ships three Agent Skills — curated, always-loadable references that shape how an AI coding agent writes validasi code, rather than a tool it has to go looking for. They live in this repo's skills/ directory, one per package, mirroring the same split as the rest of these docs.

They aren't Claude Code–specific. SKILL.md is an open format — YAML frontmatter (name, description) plus a Markdown body, with optional references/, scripts/, and assets/ subfolders — that originated at Anthropic and has since been adopted by Claude Code, Cursor, Codex CLI, GitHub Copilot, OpenCode, and dozens of other agents. The same three files here work unchanged across all of them; only where each agent looks for its skills directory differs.

Why skills, not just docs

The pages in this site are for you to read. Skills are for an agent to read — each one is a compact SKILL.md (plus deeper references/*.md files for less-common detail) written specifically to steer an AI assistant toward the correct API shape, the right rule/annotation to reach for, and the lifecycle rules (disposal, registration, async validation) that are easy to get subtly wrong from first principles. Where this site explains why something works the way it does, a skill is optimized to make the agent do the right thing by default, on the first attempt, without you having to paste in explanations.

The three skills

Skill	Covers	Companion section
validasi	Building schemas, Rules.*, transforms vs. preprocessing, custom sync/async rules, error handling	Core validasi
validasi-gen	@ValidateClass/@Validate<T> annotations, build flags and their gating rules, @RefineFn/cross-field sugar, extensibility (Inline, CustomRule, etc.)	Code Generation
validasi-ui	ValidasiFormController ownership/disposal, register/unregister and shouldUnregister, ValidasiSubmit's sync/async paths, common lifecycle footguns	Form Management

Each skill's description is written to trigger even when you don't name the package explicitly — e.g. asking an agent to "validate this form field" or "why does my controller throw after dispose" is enough for the right skill to load.

Install with npx skills

The fastest way to pull these into your own project is the open-source skills CLI (npx skills) — it works against any public GitHub repo containing skills/<name>/SKILL.md files, which is exactly this repo's layout, no extra manifest required. It also auto-detects which agent you're using and installs into that agent's convention (.claude/skills/, .agents/skills/ for Cursor/Codex/Cline/Warp/Zed, .continue/skills/, and so on) for you.

# Install one skill
npx skills add albetnov/validasi --skill validasi-ui

# Install all three
npx skills add albetnov/validasi --all

# See what's installed, or search across skill repos
npx skills list
npx skills find validasi

Manual install (no Node/npx)

If npx isn't an option, copy the relevant skill folder — e.g. skills/validasi-ui/ — from this repository into your agent's skills directory, keeping the folder name and internal structure (SKILL.md plus its references/ subfolder) intact. The destination directory depends on your agent (.claude/skills/ for Claude Code, .agents/skills/ for several others) — check your agent's docs for its exact convention. Since SKILL.md is the same format everywhere, nothing in the file itself needs rewriting.

You only need the skill(s) matching the packages you actually use — validasi-ui's skill isn't useful in a backend-only project with no Flutter forms, for instance.

Using them in this repo

If you're already working inside this monorepo, nothing to install — any agent that supports the Agent Skills format discovers skills/ on its own and triggers the matching skill based on its description whenever a task calls for it (writing a schema, wiring a form, debugging a codegen build). This isn't specific to any one agent — it's how project-level skill discovery works across the whole ecosystem.

Skills vs. the MCP server

These solve different problems, and you can use either or both:

• MCP server (validasi_mcp) is a lookup tool — your agent calls it at runtime to search or fetch this docs site when it needs to look something up. Useful for anything not baked into a skill, or for keeping up with docs changes without updating a skill.
• Skills are authored guidance — they're already loaded before the agent writes a line of code, encoding the lifecycle rules, gotchas, and "reach for X, not Y" decisions that are tedious to re-derive from a documentation lookup every time.

Using both means your agent starts from the skill's curated judgment and can still fall back to the MCP server for anything the skill doesn't cover. Both work the same way regardless of which MCP-and-skills-compatible agent you're using.