Cloudflare Agent Skills for Workers and Pages: Complete 2024 Guide

Cloudflare agent skills for Workers and Pages include nine specialized tools: agents-sdk, cloudflare, wrangler, workers-best-practices, web-perf, durable-objects, sandbox-sdk, and more—each enabling AI agents to deploy, audit, and manage edge applications programmatically.

The VoltAgent/awesome-agent-skills repository curates production-ready agent skills from major infrastructure providers. For developers building on Cloudflare's edge platform, these skills transform AI agents from passive assistants into active infrastructure operators—handling deployments, performance audits, and stateful coordination without leaving your workflow.

Cloudflare Agent Skills Available for Workers and Pages

The repository's "Skills by Cloudflare Team" section lists all Cloudflare-maintained agent skills. For Workers (serverless edge functions) and Pages (JAMstack hosting), these nine skills cover the complete development lifecycle:

Skill	Primary Purpose	Workers/Pages Application
cloudflare/agents-sdk	Stateful AI agents with scheduling, RPC, MCP servers	Run AI-driven logic inside Workers/Pages with persistent state
cloudflare/cloudflare	Comprehensive platform management	Provision Workers, Pages, KV, R2, D1, and networking via single skill
cloudflare/wrangler	Deploy and manage Workers/Pages	Canonical deployment tool for wrangler publish operations
cloudflare/workers-best-practices	Code quality auditing	AI-powered review against production standards
cloudflare/web-perf	Core Web Vitals measurement	Performance auditing for Pages sites
cloudflare/durable-objects	Stateful coordination	RPC, SQLite, and WebSocket support for complex Workers
cloudflare/sandbox-sdk	Isolated code execution	Secure sandbox environments within Workers

These skills are defined in metadata files hosted on officialskills.sh, with the repository's README.md providing curated links and descriptions.

Deploying Workers with the wrangler Skill

The cloudflare/wrangler skill exposes the full Wrangler CLI programmatically. For Cloudflare Workers and Pages deployments, this eliminates context-switching between your agent conversation and terminal.

Deploy a Worker via Agent Skill

result = await skill.invoke(
    "cloudflare/wrangler",
    {
        "command": "deploy",
        "script_path": "./src/handler.js",
        "name": "my-api-worker",
        "env": "production",
        "compatibility_date": "2024-01-15",
        "kv_namespaces": [{"binding": "CACHE", "id": "xxxx..."}],
        "d1_databases": [{"binding": "DB", "database_name": "production-db"}],
    },
)

The skill automatically handles:

• Authentication via CLOUDFLARE_API_TOKEN
• wrangler.toml generation from the JSON payload
• Streaming deployment logs back to the agent

Deploy a Pages Project

pages_deploy = await skill.invoke(
    "cloudflare/wrangler",
    {
        "command": "pages deploy",
        "project_name": "my-spa",
        "directory": "./dist",
        "branch": "main",
        "commit_hash": "abc123...",
        "commit_message": "Feature: add checkout flow",
    },
)

Auditing Workers Code with workers-best-practices

The cloudflare/workers-best-practices skill implements Cloudflare's production readiness standards. For Workers and Pages projects, this provides automated code review without manual checklist enforcement.

Run a Best-Practices Audit

review = await skill.invoke(
    "cloudflare/workers-best-practices",
    {
        "repo_url": "https://github.com/acme/edge-api",
        "branch": "main",
        "files": [
            "src/auth.js",
            "src/rate-limiter.js",
            "wrangler.toml",
        ],
        "rules": ["error-handling", "cpu-time-limits", "subrequest-limits", "secrets-management"],
    },
)

for issue in review["issues"]:
    severity = "🔴" if issue["severity"] == "error" else "🟡"
    print(f"{severity} {issue['file']}:{issue['line']} — {issue['message']}")
    if issue["suggestion"]:
        print(f"   💡 {issue['suggestion']}")

The skill checks against rules defined in Cloudflare's internal standards, including:

• Error handling: Proper try/catch and structured logging
• CPU time limits: Warning on potentially expensive operations
• Subrequest limits: Tracking fetch() calls against the 50-subrequest limit
• Secrets management: Detecting hardcoded credentials

Performance Auditing with web-perf

For Cloudflare Pages sites, the cloudflare/web-perf skill runs Core Web Vitals measurement equivalent to Lighthouse but optimized for edge-deployed applications.

Audit Pages Performance

audit = await skill.invoke(
    "cloudflare/web-perf",
    {
        "url": "https://my-app.pages.dev",
        "device": "mobile",
        "location": "LHR",  # London edge

        "metrics": ["LCP", "CLS", "FID", "TTFB", "FCP"],
        "third_party_blocking": True,
    },
)

print(f"Core Web Vitals: {'✅ Pass' if audit['passed_cwv'] else '❌ Fail'}")
for metric, value in audit["metrics"].items():
    status = "good" if value["score"] == "good" else "needs-improvement"
    print(f"  {metric}: {value['value']}{value['unit']} ({status})")

The skill runs from Cloudflare's edge network, providing realistic TTFB measurements for globally distributed applications.

Stateful Coordination with durable-objects

The cloudflare/durable-objects skill exposes Cloudflare's stateful primitives for complex Workers applications requiring persistence beyond KV's eventually-consistent model.

Create and Manage Durable Objects

# Provision a new Durable Object class

creation = await skill.invoke(
    "cloudflare/durable-objects",
    {
        "action": "create_class",
        "name": "Room",
        "class_definition": {
            "name": "ChatRoom",
            "script_path": "./src/room.js",
            "storage_backend": "sqlite",  # or "memory" for ephemeral

        },
        "bindings": {
            "websockets": True,
            "rpc_methods": ["sendMessage", "joinRoom", "leaveRoom"],
        },
    },
)

object_id = creation["object_id"]
print(f"Durable Object created: {object_id}")

For Pages with server-side logic, Durable Objects provide:

• SQLite-backed state: ACID transactions at the edge
• WebSocket coordination: Real-time collaboration features
• RPC interfaces: Type-safe method calls between Workers

Platform-Wide Management with cloudflare Skill

The cloudflare/cloudflare skill provides comprehensive platform coverage. Unlike narrow tools, this Cloudflare agent skill handles the full infrastructure stack surrounding Workers and Pages.

Provision Complete Environments

env = await skill.invoke(
    "cloudflare/cloudflare",
    {
        "project": "ecommerce-platform",
        "components": {
            "pages": {
                "name": "storefront",
                "build_command": "npm run build",
                "build_output_directory": "dist",
                "compatibility_date": "2024-01-15",
            },
            "workers": [
                {
                    "name": "api-gateway",
                    "main": "./api/index.ts",
                    "routes": ["api.example.com/*"],
                },
                {
                    "name": "webhook-processor",
                    "main": "./webhooks/index.ts",
                    "triggers": {"queues": ["payment-events"]},
                },
            ],
            "storage": {
                "kv_namespaces": [{"binding": "SESSIONS", "ttl": 86400}],
                "d1_databases": [{"binding": "ORDERS", "name": "orders-db"}],
                "r2_buckets": [{"binding": "ASSETS", "name": "product-images"}],
            },
        },
    },
)

This single invocation provisions:

• Pages site with build pipeline
• Multiple Workers with routing and queue triggers
• Storage layer (KV, D1, R2) with proper bindings

Secure Sandbox Execution with sandbox-sdk

The cloudflare/sandbox-sdk skill enables isolated code execution within Workers—critical for multi-tenant or user-generated code scenarios.

Deploy Sandbox Environment

sandbox = await skill.invoke(
    "cloudflare/sandbox-sdk",
    {
        "script": """
            export default {
                async fetch(request, env) {
                    const url = new URL(request.url);
                    const code = url.searchParams.get('code');
                    // Safely execute user code with resource limits
                    return new Response('Executed in sandbox');
                }
            }
        """,
        "runtime": "workers",
        "limits": {
            "cpu_ms": 50,
            "memory_mb": 128,
            "subrequests": 10,
        },
        "isolation": "strong",  # v8 isolates + CSP

    },
)

print(f"Sandbox URL: {sandbox['url']}")

For Pages with server-side rendering or Worker-based platforms, this provides defense-in-depth for untrusted code execution.

Summary

The VoltAgent/awesome-agent-skills repository provides nine specialized Cloudflare agent skills for Workers and Pages development:

• Deployment: wrangler skill for publishing Workers and Pages
• Code quality: workers-best-practices for automated auditing
• Performance: web-perf for Core Web Vitals measurement
• State management: durable-objects for persistent coordination
• Security: sandbox-sdk for isolated code execution
• AI agents: agents-sdk for stateful agent orchestration
• Platform-wide: cloudflare skill for comprehensive infrastructure management

These skills enable AI agents to actively manage the full lifecycle of edge applications—from provisioning infrastructure to auditing production code.

Frequently Asked Questions

What is the difference between the wrangler and cloudflare agent skills?

The cloudflare/wrangler skill focuses narrowly on deployment operations—publishing Workers, managing KV namespaces, and handling wrangler configuration. The cloudflare/cloudflare skill provides comprehensive platform access, including account management, DNS, load balancing, and infrastructure provisioning alongside Workers and Pages deployment. Use wrangler for deployment-specific workflows; use cloudflare for full-stack infrastructure management.

Can these agent skills create new Cloudflare accounts or just manage existing resources?

Based on the repository's README.md and skill definitions, the Cloudflare agent skills operate on existing authenticated accounts. They do not provision new Cloudflare accounts—you must have an active account and API token configured. The cloudflare/cloudflare skill can manage account-level resources like zones and memberships, but account creation requires manual signup through Cloudflare's website.

How does the workers-best-practices skill compare to static analysis tools like ESLint?

The cloudflare/workers-best-practices skill implements Cloudflare-specific production rules that generic tools cannot enforce. While ESLint checks JavaScript syntax and common patterns, this skill validates Workers-specific constraints: subrequest limits, CPU time boundaries, proper error handling for edge cases, and wrangler.jsonc configuration conventions. It also provides remediation suggestions tailored to Cloudflare's runtime environment.

Are these Cloudflare agent skills free to use or do they require paid plans?

The agent skills themselves are free and open-source, hosted in the VoltAgent/awesome-agent-skills repository. However, they operate against Cloudflare's platform APIs, which have their own pricing tiers. Workers and Pages have generous free tiers, but Durable Objects, KV storage, and advanced features may incur costs based on usage. The skills do not add licensing fees—they simply automate operations you could perform manually with wrangler or the Cloudflare dashboard.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}