Install
$ agentstack add skill-connorgriffin-skills-ui-mockups ✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
Verified badge
Passed review? Show it. Paste this badge into your README, it links to the public security report.
Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.
We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.
How agent discovery & health will work →About
UI mockups
Produce several genuinely different HTML mockups using the target application's real visual language, data shape, and rendering stack. Render every variant, iterate with the reviewer, and preserve one locked mockup as the implementation specification.
Use drive-local-webapp to render and exercise the mockups. If it is not installed, ask to install it from this repository before the screenshot phase. Use parallel subagents when the client supports them.
Optional integrations have explicit fallbacks:
- Use Codebase Memory for structural discovery when available; otherwise use
repository search and direct file reads.
- Use
grillingto sharpen the brief when installed; otherwise run the inline
interview in step 3.
- Use
codebase-designfor a non-trivial render module when installed;
otherwise preserve the shipping module shape and keep the logic local.
- Use
impeccablefor the final craft gate when installed; otherwise perform
the manual checks in step 7.
Choose the review mode
- Interactive mode: keep the server running, give the user the live URLs,
and wait for their choice before locking a direction.
- AgentFlow/headless mode: never wait for browser review. Render the required
states, inspect screenshots and console output, choose the direction that best satisfies the brief, and lock it. Produce the screenshots in the orchestrator's evidence/artifact directory plus mockups/.locked.md: a contract of at most 150 words stating the chosen concept, primary behavior, required states, details implementation must preserve, and evidence paths.
Ground rules
- Ground every variant.
- Read
DESIGN.mdandPRODUCT.mdwhen present, then verify their claims
against the shipping frontend.
- Lift actual light and dark theme tokens, fonts, radii, and shadows.
- Use the application's shipping UI or chart library at its shipping version.
- Prefer a deliberately manufactured fixture with the real schema. A
captured payload must come from a throwaway development/demo environment. Never collect production, private, personal, health, credential, or customer data for a mockup.
- Fork relevant render logic so the winner can be lifted into the app.
- Vary the concept, not the decoration. Change the layout metaphor,
information hierarchy, interaction model, or visualization. Three variants that differ only in color are one design.
- Mark throwaway chrome. Label variant switchers, seed controls, and other
mock-only scaffolding in comments.
- Inspect rendered output. Source review alone does not validate a visual
artifact.
Workflow
1. Establish the ledger
Read mockups/INDEX.md. Create it if absent. Record one row per surface:
| Surface | Concept | Status | Issue | File |
| --- | --- | --- | --- | --- |
Treat locked entries as binding precedent for adjacent surfaces. For shipped entries, use the application itself as ground truth rather than old mockup markup.
2. Build a grounding kit
Collect:
- the relevant screen and its primary user question;
- visual tokens and their source files;
- the UI/chart library and version;
- the shipping render module or component;
- the real data shape from a safe fixture or throwaway demo environment;
- required states such as empty, typical, dense, error, mobile, light, and dark.
Save runtime captures as mockups/.capture.json and add *.capture.json to .gitignore. Do not commit captures unless they are deliberately manufactured fixtures containing no sensitive data.
3. Fix the brief
Define one surface, one decision, hard constraints, states to render, and three or four named concept directions. A short inline interview is enough when no interviewing skill is installed.
4. Fan out
Give one fresh subagent exactly one concept. Fill in [the variant prompt](references/variant-agent-prompt.md) with the grounding kit and shared brief. Each agent writes:
mockups/-.html;mockups/--chart.jswhen rendering logic is non-trivial.
If parallel agents are unavailable, create the variants sequentially while keeping each concept brief isolated from the others.
5. Render and inspect
Serve mockups/ over HTTP, then use drive-local-webapp to render every required state. Store screenshots in a temporary directory outside the repository. Inspect the actual images and console errors.
In interactive mode, keep the server alive for review. Open the served URLs when the environment and user permit browser control; otherwise return the URLs. In AgentFlow/headless mode, save the rendered evidence and continue without opening a browser or waiting for a person.
6. Review tersely and iterate
For each variant, give one line describing its design bet and one line of judgment. Recommend one direction. Incorporate review feedback in a new render round rather than arguing from source.
7. Lock the winner
For interaction-heavy finalists, walk the primary task with two or three useful perspectives: an impatient expert, a confused first-time user, a keyboard-only user, a stress tester, or a distracted mobile user. Name the first concrete element that stalls each walkthrough and fix it.
Run an installed visual-audit skill on the finalist when available; otherwise check contrast, keyboard focus, responsive overflow, and target sizes directly. Then:
- Add a
LOCKEDheader comment to the winning HTML and companion JavaScript. - In AgentFlow/headless mode, write the required 150-word-or-shorter locked
contract and retain its evidence screenshots.
- Delete losing variants and their screenshots.
- Update
mockups/INDEX.mdwith statuslocked. - Reference the locked file from the implementation issue.
After implementation ships, set the ledger row to shipped and archive the mockup. The application then becomes the source of truth.
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: ConnorGriffin
- Source: ConnorGriffin/skills
- License: Apache-2.0
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet, be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.