From qe-framework
Guides setup of Agentation, a visual UI annotation tool that converts clicks on UI elements into structured context (CSS selectors, source paths) for AI coding agents.
How this skill is triggered — by the user, by Claude, or both
Slash command
/qe-framework:QagentationThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
This skill is a **setup and usage guide only**. It does NOT execute UI modifications.
This skill is a setup and usage guide only. It does NOT execute UI modifications.
| Request | Correct action |
|---|---|
| "Set up Agentation", "Connect visual feedback tool" | This skill — guide setup |
| "Fix the UI", "Fix the CSS", "Review the design" | NOT this skill — use Qfrontend-design, Qweb-design-guidelines, or standard code implementation |
Agentation (agentation.com) converts visual UI annotations into structured context that AI coding agents can understand. Instead of describing UI issues in text, you click directly on elements and annotate them.
| Feature | Detail |
|---|---|
| Input | Click on UI elements, write feedback annotations |
| Output | Structured markdown with CSS selectors, source paths, component tree |
| Integration | Copy-paste to agent, or MCP for direct access |
| Platform | Desktop only |
| License | Free for personal and internal company use |
| Package | agentation on npm (v2.3.3+) |
When you annotate a UI element, Agentation extracts:
Before guiding setup, check if Agentation is already available:
# Check MCP connection
claude mcp list 2>/dev/null | grep -i agentation
# Check npm package
npm list agentation 2>/dev/null || npx agentation --version 2>/dev/null
npx agentation
This injects the Agentation overlay into your running dev server.
MCP integration skips the copy-paste step. The agent directly accesses your annotations.
claude mcp add agentation -- npx agentation mcp
Or edit ~/.claude.json:
{
"mcpServers": {
"agentation": {
"command": "npx",
"args": ["agentation", "mcp"]
}
}
}
claude mcp list | grep agentation
With MCP connected, you can talk to the agent naturally:
# Resolve all feedback
"Resolve my feedback"
# Fix a specific annotation
"Fix annotation #3"
# List all annotations
"Show current annotations"
The agent can also respond back:
| Practice | Example |
|---|---|
| Be specific | "Button text is unclear - should say 'Save Draft'" not "fix this" |
| One issue per annotation | Easier for the agent to process individually |
| Include context | "Expected: 16px padding. Actual: 8px padding" |
| Use text selection | Select text for typo or content issues |
| Pause animations | Pause before annotating a specific animation frame |
# Check npm package
npm view agentation version
# Check MCP connection (if using MCP)
claude mcp list | grep agentation
npx claudepluginhub inho-team/qe-framework --plugin qe-frameworkCreates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.