Guide · Agent skill
How to use the Scientific Brainstorming agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
Scientific Brainstorming is an agent skill from K-Dense-AI's scientific-agent-skills collection that gives an AI agent a disciplined process for early-stage research ideation. It is written for researchers, students and lab facilitators who want to generate, challenge and compare candidate research directions with a visible decision trail. After this guide you will know what the skill changes about the agent's behaviour, how to install it, how to phrase requests that trigger it, and what its safety limits are.
Scientific BrainstormingK-Dense-AI/scientific-agent-skills · MIT Download the skill Quick start
- 1Download the ZIP. Download the skill ZIP from the page you are reading.
- 2Unzip the skill. Unzip it into ~/.claude/skills/ for personal use, or into .claude/skills/ inside a project. Alternatively, upload the ZIP under Settings → Capabilities → Skills in the Claude apps.
- 3Write a focal question. Ask the agent for help brainstorming and state one specific research question, its audience and the decision it supports.
- 4Generate ideas independently. Have each participant write ideas privately before any group discussion, recording assumptions and predictions for each one.
- 5Score and review shortlisted ideas. Define criteria and weights first, then score ideas and run adversarial review on the shortlist.
- 6Keep the decision log. Record decisions, dropped ideas and reasons, and run validate_register.py on the record if you use the scripts.
What is an agent skill, and what does this one make the agent do?
An agent skill is a folder containing a SKILL.md file with instructions that an compatible AI agent loads when a task matches the skill's description. This skill is designed for creating, organizing, challenging and prioritizing candidate research directions.
Once it is active, the agent follows a set of operating rules. It labels every claim as an idea, assumption, prediction, located evidence or decision. It asks for a focal question, generates ideas independently before group discussion, runs adversarial review of shortlisted ideas, and can score ideas against explicit weighted criteria. Every output is treated as a proposal, not a finding.
- Scopes a session around one focal question and records constraints and out-of-scope topics.
- Structures ideation into independent generation, sharing, clustering, criteria, review and evidence checks.
- Scores ideas with transparent weighted criteria and warns against automatic winner selection.
- Produces empty session registers or checks existing ones for structural errors.
Install the skill: download the ZIP and unzip it
Download the ZIP of the skill from the page you are reading. Then choose one of two locations depending on whether you want it available everywhere or only in one project.
For a personal install that is available across your projects, unzip the skill into ~/.claude/skills/. For a single project, unzip it into the .claude/skills/ folder inside that project. In the Claude apps you can instead upload the ZIP under Settings → Capabilities → Skills.
- Personal: ~/.claude/skills/
- Project only: .claude/skills/ inside your project folder
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
Trigger the skill with natural requests
You do not need a special command. The skill is meant to load when your request matches its description, which covers early-stage research brainstorming and prioritizing candidate directions. Describe your goal and mention the focal question, constraints and the decision you need to make.
Make clear when you want to stop at ideas. Requests that ask for validation, approval or clinical advice will be routed elsewhere by the skill.
- "Help me brainstorm research directions on microbial biofilm control for my lab group's next project."
- "Set up a brainstorming session template for six students exploring questions about soil carbon."
- "Score these five project ideas against weighted criteria for feasibility, information gain and ethics burden."
- "Run an adversarial review of my shortlisted idea and list what observation would show it is wrong."
What you need before you start
The core guidance is plain instructions and works in any host that supports Agent Skills. The optional bundled scripts require Python 3.11 or newer and use only the standard library. According to the skill's own notes, they make no network or LLM calls and need no credentials.
If you do not want to run scripts, you can still use the methods, workflows and evaluation guidance in the reference files.
- Any Agent Skills-compatible host for the core guidance.
- Python 3.11+ only if you want the bundled helper scripts.
- No network access and no credentials are required by the skill's documentation.
Use the bundled scripts for sessions, scoring and validation
The skill ships three helper scripts plus a shared module. session_scaffold.py creates a deterministic JSON starting point for a session record, and it deliberately omits a generated timestamp. evaluate_matrix.py supports weighted scoring of ideas, and validate_register.py checks records for structural errors.
The agent can call these for you, or you can run them yourself from the skill folder. Check the script's own help output for the exact options it accepts.
- scripts/session_scaffold.py: creates a starting session record.
- scripts/evaluate_matrix.py: scores ideas against weighted criteria.
- scripts/validate_register.py: checks a register for structural errors.
- scripts/_common.py: shared helper code used by the scripts.
Safety notes: scripts, data and what to hand off
The skill runs scripts, so read them before use if that matters for your setup. Per the skill's metadata, the bundled CLIs use the standard library only and make no network calls. The skill also asks the agent to record provenance without exposing confidential, personal, controlled or unpublished information.
The skill is explicit about its limits. Brainstorming cannot validate a hypothesis, and a brainstorm is never ethics, biosafety, regulatory or clinical approval. Route those topics to the relevant authorized reviewers, and route empirical design, sample size, literature review and analysis to the related skills the source names.
- Do not use the output as evidence, approval or treatment guidance.
- Do not enter confidential, personal or unpublished data into session records.
- For observation-led hypotheses, the source points to hypothesis-generation; for study architecture, experimental-design; for sample size, statistical-power; for existing evidence, literature-review; for analysis, statistical-analysis.
Tips for getting useful sessions
Write one focal question before you start, and be specific about the decision it supports. Have participants generate ideas privately before sharing, since the skill is built around that order. Define criteria and weights before anyone looks at scores, and keep disagreement visible rather than averaging it away.
Treat the weighted matrix as a decision aid. The skill warns against selecting a winner automatically, and feasibility, uncertainty and ethics checks still control the outcome.
- Set the focal question, audience and time horizon first.
- Run the independent round before literature search, then reopen ideation briefly.
- Record minority views and abstentions, not only the final shortlist.
- Keep the decision log so later reviewers can trace why ideas were kept or dropped.
What people use it for
- Generating and clustering candidate research questions for a lab group's next project.
- Running a documented brainstorming session for students with a traceable decision trail.
- Building a transparent weighted scoring matrix to compare project ideas.
- Stress-testing a shortlisted idea by listing what observation would show it is wrong.
Frequently asked questions
Is the Scientific Brainstorming skill free to use?
The skill is published under the MIT licence, as stated on its page and in its SKILL.md metadata. Check the licence file in the downloaded ZIP for the full terms.
What do I need to run the Scientific Brainstorming skill?
The core guidance works in any Agent Skills-compatible host. The optional bundled scripts require Python 3.11 or newer and use only the standard library.
Does the skill need internet access?
According to the skill's documentation, the bundled CLIs make no network or LLM calls and require no credentials. The skill's core guidance is instructional text.
Where do I install the skill in Claude?
Unzip the downloaded ZIP into ~/.claude/skills/ for personal use, or into .claude/skills/ in a project. In the Claude apps you can upload the ZIP under Settings → Capabilities → Skills.
Can this skill validate my hypothesis or approve my study?
No. The skill states that brainstorming cannot validate a hypothesis and that a brainstorm is never ethics, biosafety, regulatory or clinical approval. It routes those questions to appropriate experts or other skills.
Should I rely on the scores to pick a research idea?
No. The skill treats weighted scores as traceable decision aids and warns against automatic winner selection. Feasibility, uncertainty and ethics gates remain controlling.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
This skill suits researchers, students and lab facilitators who want a structured, traceable way to generate and compare research directions with a clear decision log. It is the wrong tool if you need hypothesis validation, study design, statistics, ethics approval or clinical guidance, which the skill itself hands off to other resources.
Download the skill