@evolver/skill-builder

Materializes skill proposals into concrete SKILL.md files via LLM generation.

Installation

pnpm add @evolver/skill-builder

Usage

import { SkillMaterializer, META_SKILL } from "@evolver/skill-builder";

// Create a skill builder (implements @evolver/core SkillBuilder interface)
const builder = new SkillMaterializer({
  model: "claude-haiku-4-5", // optional, default
});

// Build a Skill from a SkillProposal
const skill = await builder.build(proposal, parentSkills, pluginContext);
// skill: { name, trigger, content, scripts? }

// Access the meta-skill prompt for custom use
console.log(META_SKILL);

API

SkillMaterializer

Calls the Anthropic API with the META_SKILL best-practice prompt to generate SKILL.md content and optional helper scripts from a SkillProposal.

new SkillMaterializer(options?: { model?: string; client?: Anthropic })
builder.build(
  proposal:     SkillProposal,
  parentSkills: Skill[],
  context?:     PluginContext,
): Promise<Skill>
builder.parseResponse(text: string, proposal: SkillProposal): Skill

The LLM response is expected to contain:

• A ```markdown ``` block with SKILL.md content (YAML frontmatter + body)
• Optional ```typescript:filename ``` blocks for helper scripts

META_SKILL

Exported constant string containing skill authoring best practices. Covers SKILL.md structure, trigger rules, procedural instructions, content guidelines, and script conventions. Injected as system prompt during skill generation.

Configuration

| Option | Type | Default | Description | |--------|------|---------|-------------| | model | string | "claude-haiku-4-5" | Anthropic model for skill generation | | client | Anthropic | auto-created | Custom Anthropic SDK client instance |

Testing

pnpm test