1. Understand what you are installing
A skill is a folder of instructions, usually centred on SKILL.md, with optional scripts and reference files. It helps an agent follow a workflow. It does not automatically install Python packages, grant service access or validate scientific results.
2. Obtain the upstream repository
With Git installed, run the following in a terminal. Alternatively, use the repository’s Code → Download ZIP option and extract it.
git clone https://github.com/K-Dense-AI/scientific-agent-skills.git cd scientific-agent-skills
Record the commit you actually downloaded with git rev-parse HEAD. This website does not pin or certify the latest upstream revision.
3. Inspect one complete skill folder
Open skills/<skill-name>/SKILL.md. Read its referenced scripts, dependencies and permissions. Preserve the complete folder when copying it; a SKILL.md file alone may not contain everything the workflow needs.
4. Use your agent’s documented loading location
There is no universal directory for every agent. Consult your agent’s current documentation for project-level versus user-level skill directories, approval behaviour and how to verify a skill is loaded. Loading instructions does not prove that all referenced tools are available.
5. Check the environment and run a small example
Install only the dependencies required for your task, preferably in an isolated environment. Keep credentials outside tracked files. Ask the agent to identify input formats, expected outputs and verification steps before running a small sample.
Common setup problems
- Skill not detected: check the folder location, SKILL.md filename and agent-specific loading rules.
- Missing reference: copy the entire skill folder, including scripts and reference directories.
- Import or authentication error: check the documented package versions and service credentials.
- Result looks plausible but is wrong: compare against a known example and inspect intermediate outputs.
Free instructions do not imply free APIs, compute or licensed datasets.