Claude Code · Skills
Four small skills that make Claude collect before it builds, explore without committing, show you options instead of describing them, and never report a number it cannot show you the evidence for. Drop them into any repository; they work from the first message.
Install all four
Run it in your project root
for s in list-it mock-it explore synthesis; do npx -y degit toubashahzad/design-skills/.agents/skills/$s .claude/skills/$s done
Collects everything you ask for into a numbered list and builds none of it until you say go. For dictating changes faster than you want them done — the list comes back after every addition, so nothing is lost and nothing starts early.
Changes what Claude does with a request: holds it instead of acting on it.
npx -y degit toubashahzad/design-skills/.agents/skills/list-it .claude/skills/list-it
Renders two to four labelled variants inline in the chat, drawn in the project's own tokens and at the real proportions, each with a one-line trade-off. No artifact to open, no link to follow — you look, you pick.
Changes what an answer looks like: a picture rather than a paragraph describing one.
npx -y degit toubashahzad/design-skills/.agents/skills/mock-it .claude/skills/mock-it
Opens a second lane beside a running list, for working an idea out before it becomes work. Produces options and trade-offs and nothing else; an outcome joins the list only when you say so.
Changes what undecided means: a place to be unsure that does not leak into the build.
npx -y degit toubashahzad/design-skills/.agents/skills/explore .claude/skills/explore
Groups raw research responses into themes and makes every count carry the quotes behind it. No figure that is not in the data, no “most participants” standing in for a number, no quote paraphrased into something stronger than it was, and every point marked as observed or inferred. It ends by writing the findings to a dated file — sources named, limits stated — rather than leaving them in a chat nobody will scroll back through.
Changes what a finding has to prove: the evidence travels with the count, so a number you cannot check never reaches the deck.
npx -y degit toubashahzad/design-skills/.agents/skills/synthesis .claude/skills/synthesis
One more, kept here as a worked example rather than as something to install. Most of what a skill is worth saying is specific to one codebase, and that is the kind hardest to picture from a generic one.
Everything a portfolio's case-study pages need: the section model, the four caps that keep a header above the fold, the image conventions that need no code, and a script that reports what a study is currently made of and which of its pictures have no file — a thing no build will ever tell you, because a missing picture renders an empty frame and passes.
Written against one repository and only works there: its script imports that project's TypeScript by path. Read it for the shape, not to run it.
Read the source
Type /list-it to start one by hand, or just work —
each skill carries a description saying when it applies, and
Claude opens it when the moment matches.
A skill is a directory with a SKILL.md inside:
frontmatter naming it, Markdown telling Claude what to do. The
commands above drop one into
.claude/skills/.
They live in the repository, so they are version-controlled with the code and everyone who clones it gets them. Nothing is installed globally and nothing phones home.
They are plain Markdown. Open one, rewrite a rule, and the next message follows the new version — no build step, no reload.