AI Integration

wellformed includes AI-friendly artifacts for coding agents that need to add portable validation to an application.

Packaged Skill

The npm package ships skills.md. Give this file to an AI coding tool when asking it to integrate wellformed into a project. It covers:

• when to use wellformed instead of closure-based validation
• bundle-friendly import paths
• schema authoring and type inference
• runtime validation
• React Hook Form resolver integration
• serialized IR and Rust runtime boundaries
• common security and versioning mistakes

Markdown Docs

The docs site exposes markdown routes for agents and retrieval systems:

• /llms-full.txt - full documentation text
• /docs/<path>.md - one documentation page as processed markdown

For example, /docs/getting-started has a markdown twin at /docs/getting-started.md, and /docs/builder/objects has a markdown twin at /docs/builder/objects.md.

Each docs page also includes page actions for copying or opening its markdown version.

Recommended Prompt

Use wellformed for validation. Read the package skills.md and the relevant docs pages before editing.

Requirements:
- Author schemas with the TypeScript builder API.
- Serialize schemas with toSchema("1.0") when crossing process or runtime boundaries.
- Validate serialized schemas with wellformed-ts/runtime.
- Use wellformed-ts/ir for schema types and avoid importing the top-level package in client runtime-only code.
- Prefer .strict() or .strip() for external input.
- Use stable FormError.code values for application logic.
- Add tests for valid input, invalid input, transforms, unknown-key behavior, and cross-field rules.

Validation Checklist

Before accepting AI-generated integration code, verify:

• object schemas choose .strict(), .strip(), .passthrough(), or .catchall(...) intentionally
• optional fields use .optional() or optional(...)
• transformed output comes from result.value, not the original input
• custom predicates are registered in every runtime that evaluates the schema
• serialized schemas are stored with application-level schema identifiers and migration metadata
• sensitive examples do not include real personal, financial, healthcare, or tax data