skillit-core

@skillit/core

Shared types, SKILL.md renderer, and token budgeting for skillit plugins

When to Use

• You have one or more ExtractedSkill objects and need SKILL.md + references/ output
• Building a custom extraction pipeline that bypasses the TypeDoc plugin
• You need fine-grained control over rendering a single skill
• You have RenderedSkill objects from renderSkills() and need to persist them to the filesystem
• Building a custom pipeline that separates rendering from writing (e.g., for preview or dry-run)
• You need structured README sections for audit context or skill enrichment
• The audit engine calls this to check for Features and Troubleshooting sections
• You want programmatic quality feedback on JSDoc coverage before publishing skills
• Building a CI gate that blocks PRs with undocumented exports
• You have hand-written prose docs (tutorials, guides, architecture) in a docs/ directory
• You want these docs included alongside API skills for richer agent context
• You have ParsedMarkdownDoc objects from scanDocs() and need to attach them to an ExtractedSkill
• You want to include well-known root-level markdown files (ARCHITECTURE.md, MIGRATION.md, CONTRIBUTING.md) as documents
• Supplementing API skills with project-level documentation context
• You have an examples/ directory alongside source code
• You want to link example files to the exported symbols they demonstrate
• You have scanned examples with scanExamples and want to attach them to
• the relevant exported symbols inside an ExtractedSkill

Avoid when:

• Rapid local iteration where audit noise slows you down — use the skillsAudit: false option instead
• Your docs/ directory contains only auto-generated API docs — they duplicate the TypeDoc output
• Your root directory has many auto-generated or release-tracking markdown files — use explicit include lists instead
• Your examples are embedded as
• API surface: 18 functions, 25 types

NEVER

• NEVER set maxTokens below 500 — reference files become truncated mid-signature, producing broken code blocks that confuse LLMs
• NEVER pass skills with empty name — the output directory becomes a bare / path
• NEVER set maxTokens below 500 — reference files become truncated mid-signature, producing broken code blocks