academy-guide
When you ask how to do something with Claude, points you to matching Claude Academy courses, tutorials and use cases. It complements the commons catalogue, which spans several providers.
Try it after:
Skills
A skill is a short instruction file that your AI tool loads when you need it. These skills help you finish real work and get better at it as you go. Each one asks for your own idea, example or guess first, and the AI builds on it. Say "just do it" any time to skip that step.
To install them, see Start here.
The coach
-
practice-coach
A coach for learning to build with AI by improving real work.
Garden beta Say "coach me"
The Build Cycle skills
One for each step of the cycle. Call any of them by name, or let the coach pick.
-
/frame-the-problem
Turns a vague complaint or big goal into a specific problem, a desired change and a first thing to try.
Garden beta -
/brainstorm
Shows what's possible for a goal by generating many varied approaches, building on the person's own ideas first.
Garden beta -
/build-brief
Plans a small AI build before anything gets built. Picks the simplest approach that works (paper, a document or spreadsheet, a prompt, an agent or an app), writes what it must do, what can wait and how to tell if it helped, then has AI propose a plan and wait for approval of the first step.
Garden beta -
/expert-council
Vets an idea, plan, draft or decision from three to five relevant perspectives, each saying what it would do and what worries it, and ends with where they agree and disagree. Modes include a pre-mortem ("it failed — why?") and devil's advocate.
Garden beta -
/prove-it
Checks an AI result against something real before you trust it or pass it on, using the source, the actual number, or a rerun with one thing changed, and says what evidence would change the answer.
Garden beta -
/stuck-or-pivot
For when an AI build, prompt or workflow isn't working. Separates what was expected from what actually happened and what was seen from what's guessed, runs one test, then helps decide whether to fix it, make it simpler or change the approach. Can also draft a clear help request to a colleague or forum.
Garden beta -
/make-it-repeatable
Turns something that worked once into something reusable (a saved prompt, template, checklist or personal skill) plus a short share-out of what was built and learned, and can draft it as a contribution to AI Practice Commons.
Garden beta
What they practise
The skills practise six competencies. You build them by using the skills on real work, without studying them on their own.
- Check the result
- Check what AI gave you against something real before you trust it or pass it on.
- Describe the task
- Give AI the context, examples and constraints it needs to do the job well.
- Frame the outcome
- Decide what result you want, and what it's for, before asking AI for anything.
- Make it repeatable
- Turn something that worked once into a routine you, or others, can rely on.
- Steer
- Direct the work as it goes. Set guardrails, approve one step at a time, and cut scope when AI does more than you need.
- Use responsibly
- Protect private information, know when a person has to decide, and be open about how AI was used.
From the field
Open skills written by other people. The commons links to them and does not copy them. Each one is listed with credit and its licence. Install it from its own page.
discernment-nudge
After a substantive answer you might act on, adds two or three short questions that help you check key facts, probe the reasoning and notice missing context, at most once per conversation. The commons recommends keeping this light check on all the time.
Try it after:
skill-creator
Helps you turn a way of working into your own skill, and test whether it does what you meant. Use it after make-it-repeatable.
Try it after:
ChatPRD plugin (Cursor, Codex, ChatGPT)
Seven skills for writing a requirements doc from notes or code, scoring a doc on strategy, structure, clarity and completeness, turning a doc into an interactive page, and checking that what was built matches the doc. The review skill works on a pasted doc without an account; most of the rest expects a coding tool and connects to ChatPRD's own service. New, with few users so far.
Try it after:
evals-skills
For when something you built with AI runs again and again and needs real testing: reviewing failures, writing judge prompts and checking that your checks are trustworthy. The advanced step after prove-it.
Try it after:
compound-engineering (plugin)
About 36 skills for AI coding agents, built around a loop: plan, build, review, then write down what you learned so the next job can read it. For people already building software with an agent who want to make the "save what you learned" step a habit.
Try it after:
draft-review-kit
Review skills for a draft you have already written. Each reviewer has one job: check the argument, read it as a kind non-expert would, flag jargon, cut padding, or look for missing evidence and AI-sounding phrasing. A panel skill combines their notes. Use it before you send something important. One reviewer has a crude name.
Try it after:
ec-public (consulting workflows)
Four skills for everyday client work: a weekly update from material you supply, a proposal from discovery notes, email in a voice you describe, and a review of how a spreadsheet reads to someone new. The weekly update skill is a next step after prj-weekly-status-from-notes. Few people use it yet.
Try it after:
awesome-pm-skills
25 workflows that start from something you already have, such as a pricing page, interview notes or a rough feature idea, and turn it into a spec, a research summary or a test plan. Each asks up to four questions to fill gaps, and several can grade a draft you already wrote. Contains guest quotes from Lenny's Podcast that the MIT licence doesn't cover.
Try it after:
no-ai-slop
Finds the patterns that make writing sound machine-made and suggests edits that keep your meaning and tone. Use it before you send or publish anything AI helped you write.
Try it after:
human-review
Opens a page or document your AI made in the browser, where you edit the text directly and leave comments as you would in a shared doc. It then sends all your feedback to the agent at once, which is quicker and more exact than describing each change in chat.
Try it after:
fuck-cancer (patient and caregiver brief)
For someone supporting a person through cancer. It turns the reports and notes you share into a short brief you keep updating: up to three next steps, what is confirmed and what is still unclear, medical terms in plain words, care team contacts and a log of recent updates. It checks against trusted sources such as the National Cancer Institute's summaries and ClinicalTrials.gov. The brief holds health details, so decide where it lives and who sees it. It helps you prepare questions for doctors and does not replace their advice.
Try it after:
lenny-skills
About 76 product and team skills, such as writing a requirements doc, running customer interviews, prioritising a roadmap and giving feedback, each built from methods guests described on Lenny's Podcast and Newsletter. Useful once you have a real product or team task. The skills quote guests word for word; MIT covers Refound's text, not those quotes.
Try it after:
Wharton Generative AI Labs prompt library
Seventeen paste-in prompts that set AI up as a coach or role-play partner that asks one question at a time, including a devil's advocate, a team after-action review, a team charter, a meeting facilitator and a tutor. Use them when you want AI to question you instead of answering for you. A public web library, not a code repository, so there is no commit to pin.