About Grist

Why this exists

Howdy! Grist is my side quest in the world of AI. Enjoy the growing collection of prompts, skills, rules and experiments for product design, from figuring out what it can really do.

Each drop comes with a short note on why I use it and a file you can paste straight into Claude, Cursor or a CLAUDE.md. A new one lands every Friday, eight to a season.

Tested means I've used it on real work. Trying means I'm still finding out. The ideas are mine, and Claude helps me format the files. Add .md to any address if you'd rather read plain markdown.

Rarities

Rarity is how deep a drop goes. Higher tiers take more setup and give you more.

  1. CommonOne idea you can use straight away, like a rule or a short prompt.
  2. UncommonNeeds a little setup. A prompt with blanks to fill, or a checklist with steps.
  3. RareGoes deeper. A full review checklist, a longer prompt or a simple skill.
  4. EpicA template you fill in, with a worked example. Usually a spec or a bigger skill.
  5. LegendaryA whole kit to install, with examples and a note on when not to use it.
  6. UltraSeveral skills and a reviewer agent chained into one workflow. It stops for your approval along the way, and comes with a sample run.

Human-in-the-loop Checkpoints

RuleTested

tl;dr

Pause an agent before anything it can’t take back.

For deciding where an agent should stop and ask. You end up with a short list of the actions that need a person.

Why

Where a checkpoint goes depends on what the action costs. How confident the model is doesn’t matter much. I pause when an action can’t be undone, spends money or speaks for the person. Everything else, the agent can just do, as long as there’s a log people can read and an easy undo.

Teams usually add too many. When every step asks for approval, people stop reading and just click, and the one pause that mattered gets the same quick glance as the nine before it.

Rule

Human-in-the-loop Checkpoints

One line for a system prompt or CLAUDE.md

Pause when an action can’t be undone, spends money, or speaks for the person. For everything else, let the agent act, keep a readable log, and make undo easy.
human-in-the-loop-checkpoints.md

See also

  • RuleThe AI-ready Design System

    Tokens-only Guardrail

    A rule that keeps AI tools building UI from your design tokens and nothing else.

    • Tested

  • SkillThe AI-ready Design System

    Design System for Conversational AI

    Plan a design system for conversational AI products, with agent components first and names that match the code.

    • Tested

  • PromptThe AI-ready Design System

    Component Usage Guidelines

    Turn a component's code and docs into a contract that says what it's for, what it isn't for and what it promises.

    • Tested

  • PromptThe AI-ready Design System

    Token Description Lint

    Check your token names and descriptions the way a coding agent reads them.

    • Tested

I use analytics to measure and improve the website, only if you agree. Privacy policy