Practical AI "skills" for the non-profit and community sector, shared by LogicalOutcomes, a Canadian non-profit.
We share most of our tools openly because the sector is stronger when good methods are passed around. Take these, adapt them to your work, and improve them.
A skill is a plain-language instruction sheet for an AI assistant. Each one lives in its own folder and is mostly just a written document — what the task is, how to do it well, and when to use it — sometimes with a few example templates or helper files alongside.
You don't need to be technical to read one. Open the SKILL.md file in any folder here and
you'll find ordinary writing describing how an experienced person would approach the task. That
makes a skill useful in three ways: you can use it with an AI assistant, you can read it
to learn how we do the work, or you can adapt it into your own materials.
The skill format is open and not tied to any one company's AI. The format was introduced in Claude, but the same skills work with any up to date AI chat platform. Nothing here is locked to a single product. However, you may need to tweak the skills to fit your own context and model; test them and see.
The skills cover the work we do with non-profits: planning program evaluations, reviewing the research evidence behind a program, making sense of documents and interview notes, working through a tough decision from several angles, and setting up a new project. They're written for real non-profit conditions — tight budgets, grant deadlines, equity considerations, and the duty to protect people's personal information.
The table below is generated automatically from each skill's own description, so it stays
accurate as skills change. The full, authoritative description of any skill is the SKILL.md
inside its folder.
| Skill | What it does |
|---|---|
convening-experts |
Convenes expert panels for problem-solving through multi-round collaborative discussion. |
deep-background |
Meticulous fact-checking, claim analysis, and contextualization using Mike Caulfield's Deep Background methodology. |
document-analysis |
Analyze documents the user already has — evaluation reports, program documents, interview transcripts, research papers, strategic plans, data files. |
evaluation-framework-builder |
Drafts complete program evaluation plans using the LogicalOutcomes evaluation framework (Kerr & Llewelyn, 2024) — logic model, evaluation questions, indicators, interest groups analysis, project roles, schedule, and meeting agendas. |
project-starter |
Sets up new nonprofit projects using bundled handbook-style templates — drafts the project charter from a proposal, SOW, contract, email, or rough notes; assigns project roles; drafts role-and-terms emails for contractors and employees; and prepares client and internal kick-off meeting agendas. |
social-services-review |
Literature reviews and evidence syntheses for non-profit, community, social, and health services. |
Some good skills are looked after by other people. Rather than copy them here, where they would fall out of date, we keep a short, reviewed list of recommendations at
Skills/suggested-external-skills.md, with a link to the
original of each.
The simplest way — just read it. Open a skill's SKILL.md and use it as a checklist or
guide for the task yourself, or paste relevant parts into whatever AI assistant you already use,
along with your own request. This works with any assistant and needs no setup.
With an AI assistant that supports skills. Many assistants let you add a skill so it loads on its own when your request matches. Copy the skill's folder from this repository (or download the repository and use the folder you need) and add it through your assistant's skills or capabilities settings. After that, you describe your task in plain language and the assistant follows the skill — you don't have to mention it by name. The exact menu wording differs between tools, so check your assistant's help if you're unsure.
Adapt it for your organization. These are starting points, not rules. Several skills contain
blanks to fill in (like [ORGANIZATION] or a sample fee) and templates you should match to your
own policies before relying on them — especially anything about contracts, pay, or handling
personal data. Edit freely; treat your own organization's policies as the final word.
A skill helps an AI assistant do good work, but the assistant can still get things wrong. Before you rely on anything it produces, check the facts, make sure any figures and quotes are real, and confirm that sources actually exist. When you're working with real information about clients or research participants, remove personal details first — never paste people's identifiable information into a general-purpose AI assistant.
Everything here is released under the MIT licence (see the LICENSE file) — you're
free to use, change, and share it, including commercially, as long as you keep the copyright and
licence notice. Each skill also carries its own license: line in its SKILL.md, so the terms
travel with the skill no matter which tool you use it in.
Some of these build on others' work, credited in the skill's front matter and here:
convening-expertsis derived from Oskar Austegard's convening-experts skill (MIT).social-services-reviewwas partly inspired by K-Dense AI's literature-review skill (MIT).deep-backgroundbuilds on Mike Caulfield's SIFT fact-checking method (source).evaluation-framework-builderdraws on the LogicalOutcomes Evaluation Planning Handbook (Kerr & Llewelyn, 2024), which is itself shared under CC BY 4.0 (read it on SSRN).