Keep Current
Ensure responses are accurate and up-to-date by doing targeted research (use helper agents when available) before answering.
When to Research
Research when working with:
- External dependencies: Libraries, frameworks, languages, SDKs
- External services: APIs, SaaS products (WorkOS, Stripe, morphllm, etc.)
- Version-specific topics: Migration guides, breaking changes, version compatibility
- Time-sensitive topics: Recent developments, trends, "latest" features, deprecations
- Uncertain knowledge: When not confident in the recency of your answer
What NOT to Research
Skip research for:
- Language basics: Syntax, core operators, standard data structures
- Stable APIs: Well-established features that rarely change
- Universal concepts: Programming patterns, algorithms, design principles
- Confident knowledge: When certain the answer is in your knowledge base and unchanged
- General knowledge: Topics that don't depend on recent changes
Rule of thumb: If it's core programming knowledge that hasn't changed in years, don't research it.
Research Assistance
When research is needed, use available research assistance (for example, a general-purpose research agent) to gather up-to-date information:
Provide Context
Include in your prompt:
- The research question: What specific information is needed
- Context: What you're building, version requirements, language/framework
- Scope: Level of detail needed (high-level overview vs deep dive)
- Constraints: Any specific requirements or limitations
Emphasize Thoroughness
The subagent must:
- Continue researching until it has a complete, confident understanding
- Explore multiple sources (official docs, community discussions, examples)
- Verify information across sources
- Not return until it has thoroughly researched the topic
- Provide distilled findings with source URLs
Example Prompt
I need to integrate WorkOS for authentication in a Next.js app.
Research the current WorkOS documentation to understand:
- Integration setup and configuration
- SDK usage and code examples
- Authentication flow implementation
- Common patterns and best practices
Thoroughly research this topic. Do not return until you have a complete understanding of how to integrate WorkOS based on current documentation. Provide distilled findings with source URLs.
Using Subagent Results
- The subagent returns distilled, synthesized information
- Apply findings to the user's request
- Reference sources when appropriate
- If findings are insufficient, ask for more specific research
1---2name: keep-current3description: Research current online documentation and information. Use when working with external libraries, frameworks, APIs, services, version-specific issues, trends, or any topic where LLM training cutoff may impact accuracy. Guides when to use websearch, codesearch, or webfetch.4---5
6# Keep Current
7
8Ensure responses are accurate and up-to-date by doing targeted research (use helper agents when available) before answering.
9
10## When to Research
11
12Research when working with:
13
14- **External dependencies**: Libraries, frameworks, languages, SDKs
15- **External services**: APIs, SaaS products (WorkOS, Stripe, morphllm, etc.)
16- **Version-specific topics**: Migration guides, breaking changes, version compatibility
17- **Time-sensitive topics**: Recent developments, trends, "latest" features, deprecations
18- **Uncertain knowledge**: When not confident in the recency of your answer
19
20## What NOT to Research
21
22Skip research for:
23
24- **Language basics**: Syntax, core operators, standard data structures
25- **Stable APIs**: Well-established features that rarely change
26- **Universal concepts**: Programming patterns, algorithms, design principles
27- **Confident knowledge**: When certain the answer is in your knowledge base and unchanged
28- **General knowledge**: Topics that don't depend on recent changes
29
30**Rule of thumb**: If it's core programming knowledge that hasn't changed in years, don't research it.
31
32## Research Assistance
33
34When research is needed, use available research assistance (for example, a general-purpose research agent) to gather up-to-date information:
35
36### Provide Context
37
38Include in your prompt:
39- **The research question**: What specific information is needed
40- **Context**: What you're building, version requirements, language/framework
41- **Scope**: Level of detail needed (high-level overview vs deep dive)
42- **Constraints**: Any specific requirements or limitations
43
44### Emphasize Thoroughness
45
46The subagent must:
47- Continue researching until it has a **complete, confident understanding**
48- Explore multiple sources (official docs, community discussions, examples)
49- Verify information across sources
50- Not return until it has thoroughly researched the topic
51- Provide distilled findings with source URLs
52
53### Example Prompt
54
55```
56I need to integrate WorkOS for authentication in a Next.js app.
57Research the current WorkOS documentation to understand:
58- Integration setup and configuration
59- SDK usage and code examples
60- Authentication flow implementation
61- Common patterns and best practices
62
63Thoroughly research this topic. Do not return until you have a complete understanding of how to integrate WorkOS based on current documentation. Provide distilled findings with source URLs.
64```
65
66### Using Subagent Results
67
68- The subagent returns distilled, synthesized information
69- Apply findings to the user's request
70- Reference sources when appropriate
71- If findings are insufficient, ask for more specific research