Turn your knowledge into agent skills

You already hold the most valuable thing here: the wisdom in your community and your field. The eviction-response checklist. The de-escalation steps. The way your collective handles sensitive data. The tone your organization speaks in. This page is about turning that — the stuff in your head and your binders — into agent skills your AI agent follows every single time.

What’s an agent skill?

A skill is a short .md file that teaches your agent one of your methods. Instead of the agent guessing how to do something, it reads your skill and does it your way — consistently, for you and for everyone in your collective.

A harness like Goose gives the agent hands. The catalog gives it vetted tools. A skill gives it your judgment. It’s the difference between “an AI wrote something” and “an AI applied our actual playbook.”

This is the whole ethos, made concrete

We Can Just Build Things exists to augment your wisdom, not replace it. A skill is exactly that: your experience, written once, reused forever — and it stays yours.

The tool: knowledge-to-skills-pipeline

knowledge-to-skills-pipeline takes your source material — books, guides, toolkits, manuals — and turns it into structured SKILL.md files: a methodology your agent can apply, plus an attribution block that credits where the knowledge came from.

It runs on Anthropic Claude — a permitted provider (never Meta, OpenAI, or xAI), so making skills follows the same values as everything else here. Its own reference skills are drawn from Beautiful Trouble: A Toolbox for Revolution.

What goes in

Your manuals, field guides, standard operating procedures, training docs, organizing toolkits — the knowledge you’d hand a new member.

What comes out

SKILL.md files: a clear method your agent follows, with metadata and attribution so credit and licensing travel with the knowledge.

What a skill looks like

A SKILL.md is just plain text — YAML metadata at the top, then your method in plain language:

---
name: tenant-intake
description: Take an eviction report safely.
attribution:
  source: "Tenants Union field manual"
  license: CC-BY-SA-4.0
---

# Taking a tenant intake

1. Use a chosen handle, not a legal name.
2. Record the building, not the unit, at first.
3. Encrypt everything; two organizers hold keys.
4. Confirm what the tenant wants next.

The agent reads this and follows your intake process — not a generic one.

Make your first skill

1. Gather your source. A manual, a field guide, an SOP, even good notes. The clearer your source, the better the skill.
2. Run the pipeline. Follow the repo’s instructions to install and run it (it uses Bun and your own Anthropic key). It reads your material and drafts SKILL.md files.
3. Review it — you’re the expert. Read the draft. Fix anything the model got subtly wrong. Your judgment is the point; the tool just does the typing.
4. Keep the attribution honest. If the method comes from someone’s book or toolkit, credit it and respect its license (many movement resources are CC-BY-SA).

Use your skills in a build

Every starter from the Build Studio already includes a skills/ folder with a template and a README. To put your own skills to work:

1. Drop your SKILL.md files into the project’s skills/ folder.
2. Point your agent at the project (Goose or Claude Code). The generated constitution and agent prompt already tell it to read skills/ first and follow your methods.
3. Build. The agent now applies your playbook on every task — consistently, and accountable to the people who wrote it.

Your method is law, not a suggestion

Skills sit alongside the project constitution: the agent treats them as instructions to follow, not ideas to improve on. If a skill conflicts with a task, the agent should surface it and ask — never quietly do it its own way.

Browse the tool in the catalog → License, provenance, and verification for knowledge-to-skills-pipeline.