Investigate a topic responsibly and precisely. Verify sources, distinguish facts from opinions, and prioritize current information from reliable sources.
<core_rule> Use this skill for investigations that require source evaluation, synthesis across resources, or topics where official docs do not fully answer the question.
For quick API/syntax/version lookups, use lookup-docs instead.
This skill should not only return correct information — it should also help the user learn to research better: separate fact from interpretation, detect weak or outdated guidance, and decide what actually applies to their project. </core_rule>
<first_step> Before researching, establish:
Check Second Brain in Engram project second-brain if the user is researching to learn. Use Learning Sessions and Topics to understand their baseline before deciding how much scaffolding is needed.
If the topic is new and not yet applied, research directly. If the user has not said either way, ask one short clarifying question first. </first_step>
| Priority |
|---|
| Type |
|---|
| Use for |
|---|
| 1 | Official docs (MDN, NestJS, React...) | Core concepts, API, best practices |
| 2 | Academic / research papers | Deep technical comparisons |
| 3 | Established publications (O'Reilly, known experts) | Patterns, in-depth explanations |
| 4 | Community Q&A (StackOverflow, forums) | Specific problems, workarounds |
| 5 | AI-generated / mixed content | Verify elsewhere |
For community sources: always check the date, vote count, and verify against official docs before presenting to the user.
Keep these layers separate:
Present them separately when it helps avoid confusion.
Weak: Immediately recommend JWT with reasons.
Strong:
Watch for these research traps and call them out explicitly:
When useful, end research with:
Facts — directly supported by sourcesOpinion or interpretation — inferred or debatedApplies here — what fits this project or problemNext verification — what still needs checking before actingWhen the user is also learning, add:
How to read this — what clue in the source trains their research eye next timeBefore presenting the output, verify:
If any answer is no, fix it before responding.