| name | auto-research |
| description | Run a structured research loop for a topic, problem, or decision. Use when the user wants the agent to investigate a subject, compare options, map a space, or keep digging until the important questions are answered. |
Auto Research
This skill is for research that should feel agentic, but still stay grounded.
Use it for:
- topic research
- tool comparison
- market scans
- decision support
- competitive analysis
- "go deeper on this" requests
Operating Idea
Research should happen in loops:
- define the question
- map the unknowns
- search in parallel if possible
- synthesize what changed
- decide whether another loop is worth it
This is inspired by the "autoresearch" style of work:
bounded loops, better questions, tighter synthesis, then a clear next move.
Step 1: Frame The Research
Clarify:
- the main question
- the audience for the answer
- what decision or output this research should support
- how current the answer needs to be
Step 2: Build A Research Map
Write down:
- known facts
- unknowns
- assumptions
- what would count as enough evidence
Step 3: Run The First Loop
Use 3-5 targeted searches or source checks.
Aim for different lanes, for example:
- official docs
- primary announcements
- serious secondary analysis
- workflow examples from practitioners
Step 4: Synthesize
After each loop, summarize:
- what is now known
- what is still unclear
- what changed from the initial assumption
- whether another loop is justified
Step 5: Stop Intelligently
Stop when one of these is true:
- the recommendation is clear
- new searches are mostly repeating the same answer
- the remaining uncertainty does not change the decision
Output Format
When delivering the result, include:
- the answer in plain language
- the most important evidence
- open questions or risks
- the recommended next move
If the research should persist, save it into Reference/ or the relevant project folder.
