blocks
Skill1 starsChanged 5 days agoMITView on GitHub
Interactive data visuals for `show_widget` in the Claude desktop app. One set of rows feeds linked parts (filter chips, stats, tables, bars, line charts, cards, what-if sliders, step lists, send-back buttons), so a filter, a pick or a slider changes every part at once. Use before drawing a visual in chat, or when an answer holds status, numbers, findings, a comparison, a plan, a what-if or a decision the reader can explore or hand back. Not for pages meant for other people (use Artifact) or sites and apps in a repo.
3 other projects use a version of this.
Adopted by
Most recently taken up by
- SELISEdigitalplatforms/blocks-skillstheir own version, first seen Nov 8, 2026
- crissmoldovan/agent-skillstheir own version, first seen Oct 15, 2026
- dkadts/landing-kit-buildingtheir own version, first seen Aug 19, 2026
Other versions3 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 3
- SELISEdigitalplatforms/blocks-skills★ 42same name
- crissmoldovan/agent-skills★ 2same name
- dkadts/landing-kit-building★ 2same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
The file
We split this skill into the rules it gives an agent
6 rules. 0 are common, 2 are distinctive, and 4 are very specific to this project.
0
Common
Most of the 3 edited versions we read keep these.
2
Distinctive
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 14
Write the spec to a file in your scratchpad, for example findings.js.
- Distinctive: line 15
Run the builder from this skill's folder.
- Very specific: line 54
Use only numbers you were given or worked out.
- Very specific: line 58
Use act only when the next step needs you to think.
- Distinctive: line 59
Every control must change something the reader can see.
- Very specific: line 61
Keep it small: 4 stats or fewer, 8 chip values or fewer per field, 12 bars or fewer, and a limit on tables past 12 rows.
“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Blocks
A block is a small data app inside your answer. You give it rows and a list of parts. Every part reads the same rows, so when the reader filters, picks, opens or drags something, all parts change together. The point is to let the reader ask the data their own questions, which text can't do.
You write a short spec. A tested runtime draws it. You never hand-write the widget's HTML, CSS or event code.
Build one
- Write the spec to a file in your scratchpad, for example
findings.js. It holds one call:Blocks({ summary, rows, parts }). - Run the builder from this skill's folder. It checks the spec, then prints the widget code.
node <this skill's folder>/bin/blocks.mjs findings.js- Fix every
errorline and run it again. Read eachwarnline and act on it or accept it. - Pass the printed code, unchanged, as
widget_codetoshow_widget. Callread_meonce per session first if you haven't; where it differs from Blocks on look or layout, Blocks wins.
The printed code is your spec plus one script tag that loads the runtime from jsDelivr, pinned to a version. If the block shows "The Blocks runtime did not load", build again with --inline, which puts the runtime inside the widget instead.
Read spec.md for every part and option. Copy from the closest file in examples/: triage.js (filter, open, pick, send back), whatif.js (sliders and a projection), compare.js (weights re-rank options), breakdown.js (click a bar to narrow everything), plan.js (steps with status), trend.js (a chart and table that share rows).
Start from the reader's question
Pick the interaction first, then the parts. An interaction earns its place only when it changes what the reader knows or what they send back to you.
| The reader wants to ask | Give them | Parts |
| Which of these matter to me? | Narrow | facets, or bars with by |
| What's behind this one? | Open | body on a table or list |
| Where does the total come from? | Break down | two bars with by, stats, table |
| What if this number changes? | What-if | params, knobs, then derive or a series function |
| Which option wins for what I care about? | Weigh | knobs as weights, derive a score, bars and cards with best |
| What happened over time? | Read along | line, with a table when single values matter |
| Where are we in the plan? | Walk | list with rail and body |
| Do these, not those | Hand back | pick on a table, list or cards, plus act |
The runtime does the linking for you:
- A filter recounts every stat, bar, table and chart. Stats show "of" the full total while filtered.
- A changed slider shows the change from the start on stats, a "was" value on the control, and a ghost of the first line or bar on charts. Reset brings it back.
- Hovering a row or chart point marks the same row in every part. Clicking opens it.
- An
actbutton builds its message from the current picks, filters and sliders.
Rules
- The text must stand alone. Put the answer and any decision in your message, because a visual may not show everywhere. Never repeat the whole visual in words.
- Skip the block when one or two sentences say it, or when nothing in it would respond to the reader. A markdown list or table is the right tool for a short static answer.
- Use only numbers you were given or worked out. Add no currency or unit you weren't given. Put raw numbers in rows and let
fmtadd the unit. - Give the reader a sensible starting state: pre-pick your recommendation with
pick, open the row that matters withfocus, and setparamsto the real current values. - Say each fact once. Two parts may show the same rows only when each answers a different question, such as a chart for the shape and a table for exact values.
- A tone is
good,warnorbad, and only for a real state. Most rows need none. The runtime pairs every tone with an icon and words. - Use
actonly when the next step needs you to think. Filtering, sorting, sums and what-ifs run in the page. - Every control must change something the reader can see. The builder warns about picks nothing reads and sliders nothing uses.
- Parts hold data, not explanation: no titles or prose inside the block beyond labels. Keep labels to a few words, in sentence case.
- Keep it small: 4 stats or fewer, 8 chip values or fewer per field, 12 bars or fewer, and a
limiton tables past 12 rows. A chart needs enough points to show a shape; two numbers are a sentence. - For something the runtime can't draw, add a
viewpart (see spec.md) rather than hand-writing a whole widget. If the reader reports "<part> failed: ..." in the block, that text is the error to fix.
To change the runtime, add a part or run the tests, read maintain.md.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

