agent-browser-agent-builder
Skill28 starsChanged 14 days agoMITView on GitHub
Construction d'agents de navigation web autonomes incluant scraping intelligent, interaction DOM et gestion de sessions. Se déclenche avec "browser agent", "agent navigateur", "agent web autonome", "Puppeteer agent", "Playwright agent. Also triggers on "web browsing agent", "autonomous scraping agent".
- Reads credentials
- Installs packages
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 4 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
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 206
Never say "happy to help", "sure!", "great question", "let me", or similar.
- Distinctive: line 209
Stop when done. No summary, no recap, no trailing commentary.
- Distinctive: line 210
No politeness wrappers. Direct and blunt.
- Distinctive: line 213
No emoji unless asked.
≈ “Common” is an estimate: at least 1 in 5 of the 308,232 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Browser Agent Builder
1. Choisir le framework
| Critère | Playwright | Puppeteer | Selenium |
| Multi-navigateur | ✅ Chromium, Firefox, WebKit | ❌ Chrome/Edge uniquement | ✅ tous |
| Auto-wait natif | ✅ | ❌ (manuel) | ❌ (manuel) |
| Interception réseau | ✅ route() | ✅ setRequestInterception | ❌ limité |
| Support mobile | ✅ devices preset | ❌ | ❌ |
| Ecosystème / doc | Excellent (2024–2026) | Bon | Vieillissant |
Décision : Playwright par défaut sur tout nouveau projet. Puppeteer si contrainte Chrome-only ou lib existante. Selenium uniquement legacy/Java.
# Installation Playwright (Node.js)
npm init -y && npm i -D playwright
npx playwright install chromium # ou --with-deps pour CI
# Python
pip install playwright && playwright install chromium2. Architecture de l'agent
┌───────────────────────────────────────────────┐
│ Orchestrateur │
│ (LLM : planifie, décide, retry si échec) │
└────────┬──────────────┬────────────┬──────────┘
│ │ │
Navigation Extraction Mémoire/État
(goto/click/fill) (DOM/vision) (cookies, historique)Composants minimaux :
- Navigator : wraps
page.goto / click / fill / keyboard - Extractor :
page.$eval,locator, screenshot → LLM - SessionStore : persist cookies + localStorage sur disque
- DecisionLoop : LLM reçoit DOM/screenshot, émet actions structurées
3. Implémenter l'interaction DOM robuste
// Sélecteurs stables — ordre de préférence
page.getByRole('button', { name: 'Valider' }) // 1er choix
page.getByTestId('submit-btn') // 2e choix
page.getByLabel('Email') // 3e choix
page.locator('[data-id="checkout"]') // 4e choix
// Jamais : page.locator('.css-1a2b3c') // ❌ généré dynamiquement
// Auto-wait + retry inclus dans Playwright — ne pas ajouter de sleep manuel
await page.getByRole('button', { name: 'Valider' }).click()
// Shadow DOM
const shadow = page.locator('my-component').locator('pierce=button')
// iFrame
const frame = page.frameLocator('#checkout-iframe')
await frame.getByLabel('Numéro carte').fill('4111111111111111')4. Intégrer la vision LLM (multimodal)
Utiliser les screenshots quand le DOM est insuffisant (canvas, interfaces riches, CAPTCHAs visuels lisibles).
import Anthropic from '@anthropic-ai/sdk'
const client = new Anthropic()
async function describePageAndAct(page: Page): Promise<string> {
const screenshot = await page.screenshot({ type: 'png' })
const b64 = screenshot.toString('base64')
const response = await client.messages.create({
model: 'claude-sonnet-4-6',
max_tokens: 512,
messages: [{
role: 'user',
content: [
{ type: 'image', source: { type: 'base64', media_type: 'image/png', data: b64 } },
{ type: 'text', text: 'Quelle action faut-il effectuer ensuite pour compléter le formulaire ?' }
]
}]
})
return (response.content[0] as any).text
}Règle : vision = fallback coûteux — toujours tenter le sélecteur DOM d'abord.
5. Gestion de session et authentification
// Sauvegarder l'état de session après login
await page.context().storageState({ path: 'session.json' })
// Réutiliser à la session suivante
const context = await browser.newContext({
storageState: 'session.json'
})
// TOTP (2FA) avec otplib
import { totp } from 'otplib'
const code = totp.generate(process.env.TOTP_SECRET!)
await page.getByLabel('Code OTP').fill(code)6. Anti-détection — périmètre légal uniquement
// Playwright : stealth via playwright-extra
import { chromium } from 'playwright-extra'
import StealthPlugin from 'puppeteer-extra-plugin-stealth'
chromium.use(StealthPlugin())
// Profil réaliste
const context = await chromium.launchPersistentContext('', {
userAgent: 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36',
viewport: { width: 1366, height: 768 },
locale: 'fr-FR',
timezoneId: 'Africa/Tunis'
})
// Délais humains : randomiser, ne jamais fixer
const delay = (ms: number) => new Promise(r => setTimeout(r, ms))
await delay(500 + Math.random() * 1000)7. Orchestration multi-pages et error recovery
async function withRetry<T>(fn: () => Promise<T>, maxAttempts = 3): Promise<T> {
for (let i = 0; i < maxAttempts; i++) {
try {
return await fn()
} catch (err) {
if (i === maxAttempts - 1) throw err
await page.screenshot({ path: `error_attempt_${i}.png` })
await page.reload()
await delay(2000)
}
}
throw new Error('unreachable')
}
// State machine simple
type Step = 'login' | 'search' | 'extract' | 'done'
let step: Step = 'login'
while (step !== 'done') {
switch (step) {
case 'login': await doLogin(); step = 'search'; break
case 'search': await doSearch(); step = 'extract'; break
case 'extract': await extract(); step = 'done'; break
}
}8. Tests et monitoring
// Playwright Test — test E2E du workflow critique
import { test, expect } from '@playwright/test'
test('workflow checkout complet', async ({ page }) => {
await page.goto('https://shop.example.com')
await page.getByRole('button', { name: 'Ajouter au panier' }).click()
await expect(page.getByText('1 article')).toBeVisible()
})# Lancer en CI (headless, reporters JUnit)
npx playwright test --reporter=junit --output=results.xmlMonitoring : alerter si taux d'échec > 5 % sur 1 h. Versionner les snapshots DOM avec des tests de régression sur les sélecteurs clés.
Anti-patterns / Pièges
page.waitForTimeout(3000)— ne jamais utiliser de sleep fixe ; utiliserwaitForSelectorou les auto-waits Playwright.- Sélecteurs sur classes CSS générées (
css-1x2y3z) — ils changent à chaque build ; toujours utiliserrole,label,data-testid. - Pas de gestion d'erreur sur les navigations — un réseau lent ou une redirection inattendue fait planter silencieusement. Toujours
try/catch+ screenshot. - Scraper sans robots.txt check — vérifier
GET /robots.txtavant d'automatiser et respecter lesCrawl-delay. - Screenshots en fin de workflow uniquement — capturer à chaque étape critique (avant/après form submit, après login, à chaque page nouvelle).
- Ouvrir un nouveau contexte pour chaque requête — coûteux ; réutiliser
BrowserContextet changer seulement les cookies/state. - Stocker des credentials en clair dans le code — utiliser
.env+dotenv, ne jamais commitsession.json.
Bonnes pratiques 2026
- Playwright MCP (
@playwright/mcp) permet à Claude d'appeler directement les outils de navigation — idéal pour les agents LLM-driven sans coder les sélecteurs. - Préférer les
locatorchainables aux$/$$(deprecated dans les nouvelles versions). - Pour le scraping à grande échelle : Crawlee (Node) ou Scrapy + Playwright (Python) > code custom.
- Toujours isoler la logique de navigation de la logique métier pour faciliter les changements de sélecteurs sans modifier l'orchestrateur.
Communication Rules — MANDATORY
- Ultra-concise. No filler, no preamble, no pleasantries.
- Never say "happy to help", "sure!", "great question", "let me", or similar.
- Tool first, talk second. Act before explaining.
- Result first. Lead with outcome, not process.
- Stop when done. No summary, no recap, no trailing commentary.
- No politeness wrappers. Direct and blunt.
- Minimum words. If one word works, do not use ten.
- No unsolicited explanations.
- No emoji unless asked.
In this repository347 other files it gives its agents.
- agent-a2a-protocol-guideSkill
- agent-agent-observabilitySkill
- agent-api-caller-subagentSkill
- agent-autogen-guideSkill
- agent-claude-code-extension-builderSkill
- agent-code-review-subagentSkill
- agent-coding-agent-builderSkill
- agent-conflict-resolverSkill
- agent-consensus-builderSkill
- agent-context-managerSkill
- agent-cost-optimizerSkill
- agent-crewai-expertSkill
- agent-customer-support-agentSkill
- agent-data-analyst-agentSkill
- agent-database-query-subagentSkill
- agent-deployment-guideSkill
- agent-email-agent-builderSkill
- agent-evaluation-frameworkSkill
- agent-file-processor-subagentSkill
- agent-handoff-designerSkill
- agent-hierarchy-designerSkill
- agent-human-in-the-loop-designerSkill
- agent-langgraph-designerSkill
- agent-load-balancerSkill
- agent-marketplace-creatorSkill
- agent-mcp-server-builderSkill
- agent-memory-designerSkill
- agent-message-protocolSkill
- agent-monitoring-setupSkill
- agent-multi-agent-orchestratorSkill
- agent-openai-assistants-builderSkill
- agent-pipeline-composerSkill
- agent-pool-managerSkill
- agent-prompt-tunerSkill
- agent-research-agent-designerSkill
- agent-result-aggregatorSkill
- agent-retry-strategistSkill
- agent-sales-agent-builderSkill
- agent-security-hardenerSkill
- agent-semantic-kernel-guideSkill
- agent-skill-creatorSkill
- agent-skill-routerSkill
- agent-spawnerSkill
- agent-state-synchronizerSkill
- agent-subagent-delegatorSkill
- agent-supervisor-builderSkill
- agent-task-decomposerSkill
- agent-testing-frameworkSkill
- agent-tool-calling-architectSkill
- agent-voice-agent-builderSkill
- agent-web-scraper-subagentSkill
- agent-workflow-automation-agentSkill
- ai-ml-anomaly-detection-builderSkill
- ai-ml-computer-vision-guideSkill
- ai-ml-dataset-builderSkill
- ai-ml-ml-experiment-trackerSkill
- ai-ml-model-fine-tunerSkill
- ai-ml-model-optimization-guideSkill
- ai-ml-nlp-pipeline-designerSkill
- api-gateway-kong-api-gatewaySkill
- The first 60 shown.
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.

