Menu
Use when translating natural language eval instructions into playwright-cli commands during AI agent evaluation — clicking, typing, waiting, scrolling, screenshot capture
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Basé sur la classification professionnelle SOC
Interact with Langfuse and access its documentation. Use when needing to (1) query or modify Langfuse data programmatically via the CLI — traces, prompts, datasets, scores, sessions, and any other API resource, (2) look up Langfuse documentation, concepts, integration guides, or SDK usage, or (3) understand how any Langfuse feature works. This skill covers CLI-based API access (via npx) and multiple documentation retrieval methods.
Use when generating HTML evaluation reports for AI agent test suites — self-contained HTML with embedded screenshots, pass/fail results grouped by functional and compliance
Use when probing AI agents — reads eval config YAML, orchestrates browser interaction, verification, and report generation for web-based AI agent evaluation
Use when verifying AI agent responses during evaluation — content matching, behavioral assertions, AI-judged quality, and compliance checks with optional multi-step verification flows
| name | agent-eval-browser-interact |
| description | Use when translating natural language eval instructions into playwright-cli commands during AI agent evaluation — clicking, typing, waiting, scrolling, screenshot capture |
Translates natural language instructions into playwright-cli commands. Built on top of the playwright-cli skill — uses snapshots to find element refs, then acts on them.
For every natural language instruction:
playwright-cli snapshot — get current page state with element refse15)playwright-cli commandplaywright-cli screenshot --filename=<path> — capture result| Natural Language | playwright-cli Command |
|---|---|
| "Click on X" / "Click the X button" | playwright-cli click <ref> |
| "Type 'text' in X" / "Enter 'text' in X" | playwright-cli fill <ref> "text" or playwright-cli type "text" |
| "Wait for the page to load" | playwright-cli run-code "async page => await page.waitForLoadState('networkidle')" |
| "Wait for X to appear" | playwright-cli snapshot and check for element, retry |
| "Wait for the agent to respond" | See below |
| "Scroll down" | playwright-cli run-code "async page => window.scrollBy(0, 500)" |
| "Press Enter" | playwright-cli press Enter |
| "Go to URL" | playwright-cli goto URL |
# Take snapshot to see all interactive elements
playwright-cli snapshot
# Output shows refs like:
# e1 [textbox "Email"]
# e2 [textbox "Password"]
# e3 [button "Sign In"]
# e4 [link "Forgot password?"]
# Now use the ref that matches the natural language target
# "Click on the Sign In button" → playwright-cli click e3
# "Type 'admin@test.com' in the email field" → playwright-cli fill e1 "admin@test.com"
If the target element isn't obvious from the snapshot, use:
# Get more detail on a specific element
playwright-cli eval "el => el.textContent" e5
# Or take a deeper snapshot
playwright-cli snapshot --depth=6
This is the trickiest action. Strategy:
# 1. Capture current page text length
playwright-cli --raw eval "document.body.innerText.length"
# 2. (send message action happens)
# 3. Poll for new content — use run-code to wait
playwright-cli run-code "async page => {
const before = await page.locator('body').innerText();
await page.waitForFunction(
prevLen => document.body.innerText.length > prevLen,
before.length,
{ timeout: 30000 }
);
await page.waitForTimeout(2000);
}"
Also check for loading indicators:
playwright-cli snapshot
# Look for spinners, "typing..." indicators, loading states
# Wait until they disappear
After every action:
playwright-cli screenshot --filename=screenshots/step-01-click.png
If a command fails:
playwright-cli screenshot --filename=screenshots/step-N-error.pngplaywright-cli snapshot