Use when the researcher has no specific research question yet — when they need to discover what to study, not how to study it. Triggers on keywords like 'what should I research', 'research ideas', 'what can I do with this dataset', or when a dataset/paper is provided without a formed question.
Generate concrete research ideas from loose inputs — keywords, datasets, papers — and present them with actionable metadata. This skill is divergent: it expands possibilities. Its downstream partner research-brainstorming is convergent: it narrows one idea into a rigorous study design.
Core principle: You cannot refine a question you haven't asked yet. Ideation comes before design.
Use this skill when:
whats-next diagnoses the state as Pre-ideationDo NOT use this skill when:
research-brainstorminghypothesis-firstwhats-nextDiscriminating rule: If the user can state their interest as "Does/Is/Can [X] [verb] [Y]?", they have a formed question → skip to research-brainstorming. If they cannot, they are in ideation territory.
You MUST create a task for each of these items and complete them in order:
research-brainstorming handoffThe skill accepts three types of input. All optional, but at least one must be present:
Collection rules:
Adapt exploration to the input types available:
| Input type | What to do |
|---|---|
| Dataset | Read the file(s). Analyze columns, shape, dtypes, descriptive stats (N, mean, std, NaN rate, label distribution). Note what variables are available and what relationships could be studied. |
| Paper | Read abstract + methodology + limitations. Extract the research question, key findings, and identified gaps. Note what the authors suggest for future work. |
| Keywords | Web-search for recent trends, open problems, and active debates in the area. Look for survey papers, workshop topics, and recent preprints. |
When multiple input types are present, look for cross-pollination — where dataset characteristics meet literature gaps or keyword trends.
Default: 3-5 ideas. If the agent judges the scope is broad (multiple fields, large dataset with many variables, or diverse literature), expand to 5-10 ideas without asking.
Each idea should be:
research-brainstorming could start refining itEach idea follows this template:
[2-3 sentences: why this is interesting, what gap it fills]
| Item | Detail |
|---|---|
| Difficulty | High / Medium / Low |
| Data needed | Already available / Additional collection required / Public dataset available |
| Estimated duration | ~2 weeks / ~1 month / ~3 months (one researcher, full-time, excluding data collection) |
| Core methodology | e.g., "transformer + EEG temporal features" |
After presenting all ideas, recommend one:
"I find #N most promising — [one-sentence reason]. Want to start
research-brainstormingto shape this into a rigorous study design?"
The user may:
research-brainstormingresearch-brainstorming with that ideaAll are valid outcomes. The handoff is a suggestion, never a forced transition.
If the user ends the session without picking an idea:
"Want me to save these ideas before we wrap up?
research-journalcan capture them so you don't lose them."
This is a gentle suggestion. Ideas are ephemeral by default — saved only if the user journals or picks one to develop.
| What you might do wrong | What to do instead |
|---|---|
| Start designing a study for one of the ideas | Stop. That's research-brainstorming's job. Present ideas, don't refine them. |
| Generate only one idea | Always generate at least 3. The user needs options. |
| Ask 5 questions before generating | Collect inputs in one prompt, then work. |
| Force the user to pick an idea | Present, recommend, wait. The user decides. |
| Generate ideas unrelated to the inputs | Every idea must connect to at least one input (keyword, dataset feature, or literature gap). |
| Skip exploring the dataset | If a dataset is provided, you MUST look at its structure. Don't ideate in the abstract. |
eureka:using-eureka (when no formed question exists), eureka:whats-next (Pre-ideation diagnosis)eureka:research-brainstorming (when user selects an idea — suggestion only)hypothesis-first, experiment-design, or any execution/review skillFLEXIBLE — The step order (collect → explore → generate → present → recommend) is fixed and must not be skipped. But how each step executes adapts to context:
No scientific integrity is at stake at the ideation phase. The downstream rigid skills (hypothesis-first, claims-audit) enforce that discipline later.