Skills
researcher
Methodical web-search persona that finds, cross-references, and summarizes with citations.
import { createRuntime } from '@agentskit/runtime'
import { researcher } from '@agentskit/skills'
import { webSearch } from '@agentskit/tools'
const runtime = createRuntime({
adapter,
skills: [researcher],
tools: [webSearch()],
})When to reach for it
- You need sourced answers, not opinion.
- You want the agent to flag uncertainty instead of speculating.
- You're ingesting RAG context and need disciplined citation output.
Behavior
- Breaks the question into sub-queries and searches each independently.
- Cross-references across sources; flags contradictions.
- Output leads with a direct answer, then cited claims, then a confidence assessment.
Tools it expects
| Tool | Why |
|---|---|
web_search | Mandatory — required by the system prompt. |
| any RAG retriever | Optional — pairs with createRAG for internal corpora. |
Example output
Q: Main differences between PostgreSQL and MySQL for a new web app?
A: PostgreSQL excels at complex queries, JSONB, and strict SQL. MySQL is simpler to set up and faster for read-heavy simple schemas. Sources: [1] PostgreSQL docs, [2] MySQL reference manual, [3] DB-Engines comparison. Confidence: high — well-documented, stable differences.
Compose
Pair with summarizer for long reports, or critic to stress-test conclusions:
import { composeSkills } from '@agentskit/skills'
const thorough = composeSkills(researcher, critic)Related
- Skills overview · Authoring · Marketplace
- Issue #453 — researcherSkill v2 (citation-first)