Bristol AI Shed← Back to the Shed
The Shed's toolkit

Skills.

A skill is a small file of instructions that teaches your AI to do one job well, your way, every time. These are ones we actually use. Take them, change them, make them yours.

Before you install anything, from anywhere: skills run with your AI's full permissions. Read the file first, every time. Ours are short and plain English on purpose, so that takes two minutes.

Using one takes a minute

Skills run natively anywhere Claude does: Cowork, claude.ai, and Claude Code. The instructions inside work in ChatGPT and Gemini too. Pick a skill below, read it, then set it up your way:

In Cowork or claude.ai

Save the copied text as SKILL.md (or grab the zip where offered), then go to Customize → Skills → upload. That's it.

In Claude Code

Paste it into a new file: ~/.claude/skills/<skill-name>/SKILL.md and start a new session. It's live.

In ChatGPT or Gemini

The auto-loading is a Claude thing, but the instructions inside a skill are just plain prompt. Drop the text into a ChatGPT Custom GPT or Project, or a Gemini Gem, and it runs every time. Paste it at the top of a one-off chat for a quick go. You lose the automatic trigger, you keep the recipe.

Ours, shared openly

Skills from the Shed

Every card has a risk bar: the fewer segments lit, the less of your machine the skill can touch. Low risk means quick to verify, not safe to skip reading.

challenging-advisor

The challenging advisor

Disagrees with you on purpose, so you think harder.

Best for: Anyone whose AI tells them every idea is brilliant. Which is everyone.

Risk0/4

Touches nothing. Pure instructions, and that's the best score on this bar.

copy-clean

Copy clean

Strips the AI tells out of your writing so it sounds like you.

Best for: Anyone who drafts with AI and doesn't want readers to notice.

Risk2/4

reads: files you point it at · writes: edits, shown for your approval

weekly-review

Weekly review

A Friday loop that looks at how you actually worked, then tells you what to systematise.

Best for: Operators who keep doing the same thing by hand and haven't noticed yet.

Risk2/4

reads: your project folders + session history · writes: one report file

self-healing

Self-healing

A tune-up that measures what your AI setup costs, then makes it cheaper without making it worse.

Best for: Anyone whose skills and automations have grown faster than they've been pruned.

Risk2/4

reads: your skills, agents and recent session history · writes: skill edits, each shown for approval

strict-code-review

Strict code review

An unusually tough second pair of eyes on what you build.

Best for: Builders who suspect 'it works' is not the same as 'it's good'.

Risk1/4

reads: your branch's changes

magnetic-influence

Magnetic influence

A negotiation and influence coach built on Tim Castle's MAGNETIC framework. Scores your draft or deal against eight principles, then hands you the line to say.

By Tim Castle · the book

Best for: Preparing a negotiation, pitch, pricing conversation or high-stakes email. Pairs with the book.

Risk0/4

Touches nothing. Pure instructions, and that's the best score on this bar.

Multi-file skill: upload the zip straight into Cowork or claude.ai (Customize → Skills), or unzip into ~/.claude/skills/ for Claude Code.

Not ours, still good

Worth knowing about

The best skill is one you wrote

Build your own

Build your own in three steps

  1. 01 Copy the template below into a new file: ~/.claude/skills/your-skill-name/SKILL.md
  2. 02 Fill it in. Plain English works. Write it like you'd brief a sharp new colleague.
  3. 03 Start a new session and use it. If it doesn't behave, sharpen the wording and go again.

Built something useful?

Bring it to the next Shed and show it, or email it to hello@bristolaished.com. If it's good, it goes up here with your name on it. That's the whole deal: take what's useful, share the next thing.