The idea in one sentence
An AI skill is a reusable set of instructions that gives an AI agent a method for a particular job, packaged in a folder containing a SKILL.md file. Agent Skills overview
Think of it as a playbook you can inspect and edit, not a new AI model or a substitute for a colleague.
Why use one for nonprofit work?
A useful playbook gives a task more structure. Instead of asking only for “a grant proposal,” you can work from a repeatable process: gather the funder's requirements, identify missing evidence, draft the narrative, and check the budget against the request.
For example, this library's Grant writing skill covers letters of inquiry, proposals, budget narratives, and grant reports; Grant research handles finding and assessing funding opportunities.
Start with one task you already understand. Compare the result with your normal process, and keep what is useful.
What is inside?
The format combines a small metadata block, called YAML frontmatter, with readable Markdown instructions; the wider standard also allows supporting scripts, references, and assets. Agent Skills specification
In this library, each skill is a standalone SKILL.md with a name, description, MIT license field, supervision level, and a note explaining that level. Library on GitHub
The body provides the working method: task boundaries, frameworks, deliverables, steps, and pitfalls, with practitioner or advisor guidance where relevant. Library on GitHub
You do not need to understand YAML to read a skill. Begin with its description and instructions before deciding whether to install it.
A prompt, a skill, and an agent do different jobs
- Your prompt sets today's assignment. Include the audience, goal, constraints, and material the agent may use.
- The skill supplies a reusable method. Compatible agents can discover a skill from its description and load its full instructions when selected. Agent Skills overview
- The agent does the work using its available tools. Installing instructions does not by itself connect your donor database, grant portal, or financial system.
Do not assume that mentioning a topic guarantees the right skill will load. Check the product's available skills and explicitly select the intended skill when possible; the installation guide explains how.
Read the supervision label first
The September 9, 2026 library snapshot contains 63 skills across 11 categories, with 7 unsupervised, 38 review, and 18 expert-required labels. Library on GitHub
| Label | How to use the output |
|---|---|
unsupervised |
Use ordinary editing and fact-checking. The label does not mean error-free. |
review |
Have a knowledgeable staff member review it before use or circulation. |
expert-required |
Have an appropriately credentialed professional review it before filing, adoption, or reliance. |
These labels describe the library's intended supervision levels, not an automatic approval system in your AI tool. Library supervision guidance
For every level, protect confidential information, verify important facts, and keep a person responsible for the final decision.
Try one
Use Pick the right skill to match a task to a playbook, then follow Install and use per agent.