Skip to content
Back to skills

Jsgui3 Lab Experimentation

ASecurity

Use repo lab experiments (src/ui/lab) to answer jsgui3 “how should we do this?” questions, then distill the validated result into durable docs + Skills.

  • 666 stars
  • 0 votes
  • 0 copies
  • 4 views
  • Added February 9, 2026
researchnodeperformance

Works with

  • cli

Security analysis

A100/100

Scanned February 12, 2026

npx -y skills add majiayu000/claude-skill-registry --skill jsgui3-lab-experimentation --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Jsgui3 Lab Experimentation?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Jsgui3 Lab Experimentation
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/majiayu000-jsgui3-lab-experimentation/badge)](https://www.skillsdirectory.com/skills/majiayu000-jsgui3-lab-experimentation)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: jsgui3-lab-experimentation
description: Use repo lab experiments (src/ui/lab) to answer jsgui3 “how should we do this?” questions, then distill the validated result into durable docs + Skills.
---

# jsgui3 Lab Experimentation

## Scope

- Run existing lab experiments to confirm behavior (event delegation, activation, MVVM, mixins, helpers)
- Create minimal new lab experiments when encountering unknown or disputed jsgui3 behavior
- Promote stable findings into the repo memory system (Skills + Patterns/Lessons + guides when appropriate)

## Inputs

- The question you’re trying to answer (1 sentence)
- The target surface (control/server route/activation path/event semantics/perf)
- Whether you need SSR-only, client activation, or browser semantics (delegation/bubbling)

## Procedure

1. Inventory prior art
   - Search docs first: `node tools/dev/md-scan.js --guide --search "<topic>" --json`
   - Then search sessions: `node tools/dev/md-scan.js --dir docs/sessions --search "<topic>" --json`

2. Check whether a lab experiment already covers it
   - Open lab index: `src/ui/lab/README.md`
   - Open manifest: `src/ui/lab/manifest.json`
   - Optional: render the Lab Console card list (quick sanity): `node src/ui/lab/checks/labConsole.check.js`

3. Run the smallest relevant check(s)
   - Each experiment should have a `check.js` (or `check.all.js`) that exits cleanly.
   - Prefer targeted runs, e.g.:
     - `node src/ui/lab/experiments/002-platform-helpers/check.js`
     - `node src/ui/lab/experiments/004-theme-mixin/check.js`

4. If the answer is still unclear, create a minimal new experiment
   - Create `src/ui/lab/experiments/NNN-short-slug/`
   - Add `README.md` (hypothesis + expected result) and `check.js` (deterministic assertions)
   - Add an entry to `src/ui/lab/manifest.json` and the table in `src/ui/lab/README.md`

5. If the behavior touches event semantics, run the delegation suite
   - Use the shared runner (reuses one browser/page):
     - `node src/ui/lab/experiments/run-delegation-suite.js --scenario=005,006`
   - Add/adjust scenarios if your change affects capture/bubble/selector matching/stopPropagation.

6. If the behavior requires SSR + client activation + real interactivity, run a Puppeteer scenario suite
    - Use the scenario suite runner (single browser, many scenarios):
       - `node tools/dev/ui-scenario-suite.js --suite=scripts/ui/scenarios/control-harness-counter.suite.js`
    - This is the recommended fast path for validating control activation wiring (DOM linking, event listeners, state updates)
       without paying the cost of full Jest/Puppeteer E2E per test.
    - Prefer this over adding more “JS-only DOM semantics” scenarios when what you really need is jsgui3 activation behavior.

7. Distill the validated finding into durable memory
   - Create/extend a Skill (this registry): `docs/agi/SKILLS.md`
   - Add/extend a Pattern if it’s broadly reusable: `docs/agi/PATTERNS.md`
   - Write a guide if it’s a subsystem-level discovery: `docs/guides/…`
   - Record evidence + commands in the current session directory under `docs/sessions/...`

## Validation

- Lab console renders: `node src/ui/lab/checks/labConsole.check.js`
- Run the specific experiment check(s) you relied on (examples):
  - `node src/ui/lab/experiments/001-color-palette/check.js`
  - `node src/ui/lab/experiments/002-platform-helpers/check.js`
  - `node src/ui/lab/experiments/004-theme-mixin/check.js`
- If delegation/event semantics are involved:
  - `node src/ui/lab/experiments/run-delegation-suite.js --scenario=005,006,007,008,011,014`

- If you need SSR + activation + browser interaction coverage:
   - `node tools/dev/ui-scenario-suite.js --suite=scripts/ui/scenarios/control-harness-counter.suite.js`

## Escalation / Research request

Ask for dedicated research if:

- A behavior only reproduces with SSR + activation + bundling and no existing lab harness covers it
- A change appears to require new shared harness utilities (browserHarness / activation harness)
- The experiment needs performance instrumentation (counts, timings, memory) beyond a simple check

## References

- Lab index: `src/ui/lab/README.md`
- Lab manifest: `src/ui/lab/manifest.json`
- Delegation suite runner: `src/ui/lab/experiments/run-delegation-suite.js`
- Puppeteer scenario suites guide: `docs/guides/PUPPETEER_SCENARIO_SUITES.md`
- Pattern: “Lab-First jsgui3 Knowledge Gap Closure” in `docs/agi/PATTERNS.md`
 - General experimental research SOP: `docs/agi/skills/experimental-research-metacognition/SKILL.md`

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.