@agentskit/skills — for agents

#Install

npm install @agentskit/skills

#Primary exports

#Ready-made skills

• researcher — methodical web researcher.
• coder — TDD-first coder.
• planner — step-by-step planner.
• critic — critic / reviewer.
• summarizer — structured summarizer.
• codeReviewer — PR review with severity tags.
• sqlGen — NL → parameterized SQL.
• dataAnalyst — hypothesis-driven data analysis.
• translator / translatorWithGlossary — faithful translator (+ glossary variant).
• prReviewer — opinionated PR reviewer.
• sqlAnalyst — SQL analysis + recommendations.
• technicalWriter — technical writing assistant.
• securityAuditor — security review with OWASP framing.
• customerSupport — empathetic support agent.

#Vertical skills (regulated domains)

• healthcareAssistant — refuses diagnosis / dosage / triage.
• clinicalNoteSummarizer — SOAP-format summarization, never interprets.
• financialAdvisor — refuses tickers / "should you" / payment decisions.
• transactionTriage — bookkeeping triage, fixed-shape output.
• legalAssistant — informational legal explainer; never gives legal advice.
• contractReviewer — flags risky clauses with severity tags; not a substitute for counsel.
• tutor — Socratic tutor; questions/hints first, direct answers only on explicit opt-out.
• curriculumDesigner — lesson plans + rubrics tagged to Bloom levels.
• storefrontConcierge — e-commerce concierge for product discovery + cart.
• merchandisingAnalyst — e-commerce merchandising / SKU performance analyst.
• listingConcierge — real-estate listing concierge for buyers / renters.
• marketAnalyst — real-estate market analyst (comps, trends, neighbourhood).

#Composition + discovery

• composeSkills(a, b, ...) — validate and defensively merge skills into one; composed names satisfy ADR 0005 S1.
• getBuiltinSkills() — defensive full definitions for all 26 bundled skills.
• listSkills() — metadata for every bundled skill.

#Marketplace

• createSkillRegistry(initial?) — publish / list / install / unpublish.
• parseSemver, compareSemver, matchesRange — strict SemVer helpers for exact, ^, ~, >=, and * ranges; prereleases are excluded unless explicitly targeted.
• Types: SkillPackage, SkillRegistry, SkillRegistryQuery.

See Skill marketplace.

tools and delegates inside a skill are declarative names. The runtime does not gain those capabilities unless the caller supplies matching tool and skill registries. onActivate is the only executable field and is reserved for constructing per-user or per-tenant dynamic tools.

The package remains beta. Its implementation is being prepared for the stable surface proposed in RFC 0009, but promotion still requires the release history and soak evidence mandated by ADR 0024.

#Minimal example

import { createRuntime } from '@agentskit/runtime'
import { researcher } from '@agentskit/skills'

const runtime = createRuntime({
  adapter,
  systemPrompt: researcher.systemPrompt,
  tools: [/* webSearch() etc. */],
})

#Related

• @agentskit/core — SkillDefinition contract.
• @agentskit/runtime.

#Source

• npm: https://www.npmjs.com/package/@agentskit/skills
• repo: https://github.com/AgentsKit-io/agentskit/tree/main/packages/skills