validation
Skill1 starsChanged 3 months agoMITView on GitHub
Try to falsify the design before building it — name what must be true, pick the cheapest test that could prove it false, run it. Use after flows and content exist and before visual or code investment, when the team is sure users will get it, or when two design options are being argued from opinion.
20 other projects use a version of this.
Adopted by
Most recently taken up by
- thistleknot/skillstheir own version, first seen Nov 8, 2026
- GoogleCloudPlatform/cloud-solutionstheir own version, first seen Nov 8, 2026
- LeanMeanCleanApps/CleanModelContextProtocoltheir own version, first seen Nov 8, 2026
- AnyOrgName/php-skillstheir own version, first seen Nov 8, 2026
- ruujOnChain/HedgeFund-Skillstheir own version, first seen Nov 8, 2026
- honeydew-ai/honeydew-ai-coding-agents-pluginstheir own version, first seen Nov 1, 2026
Used with
Projects with it also use testing.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
NVIDIASkilltesting
Run Python and Rust tests for physicsnemo-curator using uv + pytest with coverage reporting via pytest-cov, and cargo-nextest for Rust tests. Includes benchmark workflows with pytest-benchmark and criterion.
Also in 3 projects that use this
Other versions20 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 20
- openai/codex-security★ 11ksame name
- openai/plugins★ 7.8ksame name
- GoogleCloudPlatform/cloud-solutions★ 139same name
- richard-kim-79/archora-skills★ 49same name
- honeydew-ai/honeydew-ai-coding-agents-plugins★ 43same name
- wolverin0/claude-skills★ 41same name
- loocor/mcpmate★ 35same name
- ruujOnChain/HedgeFund-Skills★ 17same name
- aibtcdev/skills★ 11same name
- AnyOrgName/php-skills★ 4same name
10 more not listed
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.
In this repository14 other files it gives its agents.
- accessibilitySkill
- colorSkill
- content-designSkill
- design-reviewSkill
- design-systemSkill
- implementation-reviewSkill
- interaction-designSkill
- layoutSkill
- measurementSkill
- motionSkill
- product-evidenceSkill
- product-framingSkill
- typographySkill
- visual-directionSkill
The file
We split this skill into the rules it gives an agent
1 rule. 0 are common, 1 is distinctive, and 0 are very specific to this project.
0
Common
Most of the 20 edited versions we read keep these.
1
Distinctive
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 34
Test the prototype with real copy: lorem ipsum invalidates comprehension results entirely.
“Common” is measured: kept by at least 6 in 10 of the 20 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
Validation
The designer produces a hypothesis; the test produces information. This skill's question is never "is this design good?" It is: what must be true for this solution to work, and how cheaply can we show it false? A design that survives a real attempt to kill it earns the visual and code investment. One that was never attacked just hasn't failed yet.
Extract the load-bearing assumptions
From the frame, the flows, and the content, list what must be true, in the form "users will/can/understand X". Then rank by damage if false × cheapness to test. You test the top three, not all of them. Examples of load-bearing:
- Users understand what the primary action does from its label alone.
- Users can find the entry point without being shown.
- The mental model matches: users expect an "Invoice" to be editable after sending, or they don't.
- Users will tolerate the required input (the form, the permission grant, the wait).
The test menu — cheapest that can falsify
| Test | Kills the assumption that... | Cost |
| Comprehension test | The words are understood. Show the screen 5 seconds, ask "what does this do, what would you click first?" | An hour, 5 people |
| First-click test | The entry point is findable. One task, one static screen, where do they click? | An hour |
| Task completion on prototype | The flow works end to end. Clickable wireframe, real labels, one task, count completions and wrong turns | Half a day |
| Edge-case walkthrough | The design survives reality. Feed it the declared state matrix: empty, partial, error, permission-denied, 10x data | An hour, no users needed |
| Wizard-of-Oz | The value is wanted at all. Fake the feature manually behind a real button | Days, reserve for expensive bets |
Five users find most comprehension and findability failures. Statistical significance is for measurement after ship, not for this stage.
Rules of the game
- Decide the kill condition before running: "if fewer than 4 of 5 find the entry point, the navigation changes". A test whose result can't change the design is theater.
- Test the prototype with real copy: lorem ipsum invalidates comprehension results entirely.
- Watch what people do; discount what they predict ("I would definitely use this" carries zero information).
- A failed test is the system working. Route the finding backward: wrong words to content-design, wrong flow to interaction-design, wrong problem all the way back to the frame.
- Findings enter the evidence ledger as facts or signals with the test as
source:. That's the loop: tested design becomes evidence for the next frame.
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.