<skill_overview> Optimize use of "assist/research/web-research" subagent for information gathering

Need current API/library documentation Researching best practices and patterns Finding solutions to known problems Searching code examples and tutorials

</skill_overview> <task_formulation>

Be specific about technology and version One request = one topic/goal (1–3 subpoints max) State expected output format Include context about intended use Specify depth level when needed: standard, deep, expert Multi-stage research is orchestrated by the caller (separate requests per stage)

Find current Next.js 14 App Router documentation: server components patterns, data fetching, and caching strategies Specific version, clear topics, actionable scope

Research Zustand vs Jotai for React state management: performance benchmarks, TypeScript support, bundle size comparison Comparative, measurable criteria, focused scope

Learn about React Too broad, no specific goal or deliverable

Split a wide request into separate topics: (1) Run CLIProxy in docker/compose, (2) Configure multiple instances, (3) Add accounts/credentials One topic per request; caller coordinates multi-step research

</task_formulation> <research_strategies>

Broad landscape scan Detailed analysis of selected options Edge cases and real-world examples

Overview of modern React state management solutions in 2025 (separate request) Deep dive into Zustand: architecture, performance patterns, TypeScript best practices (separate request) Find production Zustand examples: large-scale apps, testing patterns (separate request)

Multi-stage research must be orchestrated by the caller as separate requests (one topic per request)

Compare [A] vs [B] for [use-case]: [criteria-1], [criteria-2], [criteria-3]. Include recent benchmarks.

Find solutions for [specific error/issue] in [technology] [version]. Include workarounds and root cause.

</research_strategies> <agent_limitations> Access local project files or codebase Edit or write files Remember previous sessions </agent_limitations> <depth_levels> Summary + key sources + essential facts Comparison, edge cases, pros/cons, multiple sources Implementation details, pitfalls, validated code examples </depth_levels>