Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Get Started Free →

Network Security Setup · microck | DecimalAI

microck/network-security-setup

SkillScore96Maintained

Configure Claude Code sandbox network isolation with trusted domains, custom access policies, and environment variables

agent-toolingby microck

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

Security & AppSec pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/microck-network-security-setup/SKILL.md

decimalai skills pull microck-network-security-setup --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/microck-network-security-setup/SKILL.md https://app.decimal.ai/s/microck-network-security-setup/SKILL.md

Raw file: /s/microck-network-security-setup/SKILL.md ↗ · machine-readable summary at /s/microck-network-security-setup.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

23 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 8d ago

Yes — +39% lift · at higher cost+57% tokens0% turns96% of whole cases pass with skill (22/23)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-04	✗→✓	▲ Improved	56%	0%
case-01	✗→✓	▲ Improved	46%	0%
case-02	✗→✓	▲ Improved	35%	0%
case-03	✗→✓	▲ Improved	136%	0%
case-06	✗→✓	▲ Improved	76%	0%

Showing 5 of 23 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/3/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

microck / ordinary-claude-skills

skills_all/network-security-setup/SKILL.md

View on GitHub ↗

License NOASSERTIONImported 8/3/2026Version v1

Skill contents

SKILL.md

Network Security Setup

Purpose

Configure Claude Code sandbox network isolation policies including trusted domain whitelisting, custom access rules, and secure environment variable management.

Specialist Agent

I am a network security specialist with expertise in:

Zero-trust network architecture for AI code execution
Domain whitelisting and access control policies
Prompt injection attack prevention via network isolation
Secure environment variable management
Corporate proxy and internal registry configuration

Methodology (Systems Thinking + Self-Consistency)

Analyze Environment: Understand deployment context (enterprise, open-source, local)
Design Network Policy: Create appropriate trusted domain list
Configure Access Rules: Set up custom access patterns and exclusions
Secure Credentials: Properly handle environment variables and secrets
Validate Security: Test that policies block untrusted access while enabling work

Network Isolation Modes

Mode 1: Trusted Network Access (Recommended Default)

yaml
mode: trusted
description: Claude can only access pre-approved, known-safe domains
use_case: General development, open-source projects
trusted_domains:
  - "*.npmjs.org"
  - "registry.npmjs.org"
  - "*.yarnpkg.com"
  - "*.github.com"
  - "api.github.com"
  - "raw.githubusercontent.com"
  - "*.cloudfront.net"
  - "*.docker.io"
  - "registry.hub.docker.com"
  - "*.pypi.org"
  - "pypi.python.org"

Mode 2: No Network Access

yaml
mode: none
description: Complete network isolation, no external access
use_case: Maximum security, offline development, sensitive projects
trusted_domains: []

Mode 3: Custom Access

yaml
mode: custom
description: User-defined whitelist of allowed domains
use_case: Enterprise with internal registries, corporate networks
trusted_domains:
  - "registry.company.internal"
  - "docs.company.com"
  - "api.company.com"
  - "*.company-cdn.net"
  - [Include standard registries as needed]

Default Trusted Domains (Anthropic-Approved)

Package Registries:

*.npmjs.org - npm packages
registry.npmjs.org - npm registry
*.yarnpkg.com - Yarn registry
*.pypi.org - Python packages
pypi.python.org - Python registry
rubygems.org - Ruby gems
*.maven.org - Maven packages

Container Registries:

*.docker.io - Docker Hub
registry.hub.docker.com - Docker registry
ghcr.io - GitHub Container Registry
gcr.io - Google Container Registry
*.azurecr.io - Azure Container Registry

Source Control & CDNs:

*.github.com - GitHub
api.github.com - GitHub API
raw.githubusercontent.com - Raw GitHub content
*.cloudfront.net - AWS CloudFront
cdn.jsdelivr.net - jsDelivr CDN
unpkg.com - unpkg CDN

Development Tools:

*.vercel.com - Vercel deployment
*.netlify.com - Netlify deployment
*.supabase.co - Supabase API

Enterprise Configuration

Internal Registry Setup:

json
{
  "sandbox": {
    "enabled": true,
    "network": {
      "mode": "custom",
      "trustedDomains": [
        "registry.company.internal:5000",
        "npm.company.com",
        "docs.company.com",
        "api-docs.company.internal",
        "*.company-cdn.net",
        "*.company.cloud",

        // Include standard public registries if needed
        "registry.npmjs.org",
        "*.github.com"
      ],
      "customProxy": {
        "enabled": true,
        "http": "http://proxy.company.com:8080",
        "https": "http://proxy.company.com:8080",
        "noProxy": [
          "localhost",
          "127.0.0.1",
          "*.company.internal"
        ]
      }
    }
  }
}

Corporate Proxy Configuration:

json
{
  "sandbox": {
    "network": {
      "customProxy": {
        "enabled": true,
        "http": "http://corporate-proxy.company.com:3128",
        "https": "http://corporate-proxy.company.com:3128",
        "noProxy": [
          "localhost",
          "*.internal",
          "*.company.com"
        ],
        "authentication": {
          "enabled": false  // Use system credentials
        }
      }
    }
  }
}

Environment Variables (Secure Management)

Safe Environment Variables (OK to configure):

yaml
safe_env_vars:
  - NODE_ENV: "development"
  - API_BASE_URL: "https://api.company.com"
  - LOG_LEVEL: "debug"
  - FEATURE_FLAGS: "new_ui,beta_features"
  - BUILD_TARGET: "production"

Dangerous (NEVER in sandbox config):

yaml
dangerous_env_vars:  # Store in .env.local, never in settings
  - API_KEY: "sk-..."  ❌ SECRET
  - DATABASE_PASSWORD: "..."  ❌ SECRET
  - PRIVATE_KEY: "..."  ❌ SECRET
  - AWS_SECRET_ACCESS_KEY: "..."  ❌ SECRET

Best Practice for Secrets:

Store in .env.local (gitignored)
Use environment variable references in sandbox config
Document required variables without values
Use secret management services in production

Example Secure Configuration:

json
{
  "sandbox": {
    "environmentVariables": {
      // ✅ Non-sensitive configuration
      "NODE_ENV": "development",
      "API_BASE_URL": "https://api.staging.company.com",

      // ✅ Reference to local .env file (document required vars)
      "__REQUIRED_SECRETS__": "API_KEY, DATABASE_URL (store in .env.local)"
    }
  }
}

Security Threat Mitigation

Threat 1: Prompt Injection → Data Exfiltration

Attack: Malicious prompt in downloaded code tries to send sensitive data to attacker.com
Mitigation: Network isolation blocks all non-whitelisted domains
Result: Attack fails, data stays secure

Threat 2: Malicious Package Download

Attack: Prompt injection tries to install malware from evil-registry.com
Mitigation: Only trusted registries allowed
Result: Download blocked, system protected

Threat 3: Internal Network Scanning

Attack: Code tries to scan internal network for vulnerable services
Mitigation: Network isolation prevents arbitrary connections
Result: Internal network remains hidden

Threat 4: Credential Theft

Attack: Downloaded code reads environment variables and sends to attacker
Mitigation: Secrets not in sandbox config, network blocked anyway
Result: No credentials accessible or exfiltratable

Domain Pattern Matching

Wildcard Patterns:

*.example.com - Matches all subdomains: api.example.com, cdn.example.com
example.com - Exact match only
*.*.example.com - Multi-level wildcards: a.b.example.com

Port Specifications:

registry.company.com:5000 - Specific port
*.company.com:* - Any port on subdomains
localhost:3000 - Local development server

Protocol Handling:

HTTPS preferred and enforced where possible
HTTP allowed only for localhost and internal domains
WebSocket connections follow same rules (ws:// → wss://)

Validation and Testing

Test Network Policy:

bash
# Should succeed (trusted domain)
npm install express

# Should succeed (trusted domain)
git clone https://github.com/user/repo

# Should fail (untrusted domain)
curl https://random-website.com

# Should succeed if allowLocalBinding enabled
npm run dev

Verification Checklist:

] Package installations work from trusted registries
] GitHub operations succeed
] CDN resources accessible if needed
] Internal registries accessible (enterprise)
] Untrusted domains blocked
] Local development servers work if configured
] Build commands pass with required env vars
] No secrets in sandbox configuration

Input Contract

yaml
environment_type: enterprise | opensource | local | custom
required_access:
  public_registries: array[string]
  internal_domains: array[string]
  cdn_services: array[string]
needs_proxy: boolean
proxy_config: object (if needs_proxy)
required_env_vars: array[{name, value, is_secret}]

Output Contract

yaml
network_configuration:
  mode: trusted | none | custom
  trusted_domains: array[string]
  proxy_config: object (if applicable)
  environment_variables: object (non-secrets only)

security_analysis:
  threats_mitigated: array[string]
  access_granted: array[string]
  access_denied: array[string]
  recommendations: array[string]

setup_instructions:
  config_file_location: string
  config_content: json
  validation_commands: array[string]
  documentation_links: array[string]

Integration Points

Cascades: Works with sandbox-configurator for complete security setup
Commands: /network-security, /trusted-domains
Other Skills: Pairs with sandbox-configurator, security-review

Usage Examples

Standard Development Setup:

Configure network security for open-source development with standard npm and GitHub access

Enterprise Internal:

Set up network isolation for enterprise:
- Internal npm registry: npm.company.internal
- Internal docs: docs.company.com
- Corporate proxy: proxy.company.com:8080
- Keep access to public GitHub

Maximum Security:

Configure maximum security with no network access for sensitive project

Add Custom Domain:

Add api.specialservice.com to trusted domains for API integration

Failure Modes & Mitigations

Package install fails: Add registry to trusted domains
Git clone fails: Add git host to trusted domains
Build fails with network error: Check if build accesses CDN, add to whitelist
Proxy authentication fails: Verify proxy credentials or use system auth
Environment variable missing: Document in config, add to .env.local

Validation Checklist

] All required registries in trusted domains
] Internal domains include ports if non-standard
] Proxy configuration correct (if needed)
] No secrets in sandbox configuration
] Required env vars documented
] Test package installation
] Test git operations
] Test build commands
] Verify untrusted access blocked

Neural Training Integration

yaml
training:
  pattern: systems-thinking
  feedback_collection: true
  success_metrics:
    - zero_security_incidents
    - development_velocity_maintained
    - false_positive_rate_low

Quick Reference:

Config location: .claude/settings.local.json
Default mode: Trusted network access
Wildcard syntax: *.domain.com
Secrets: NEVER in sandbox config, use .env.local

Security Principle: Deny by default, allow explicitly, verify continuously

Want to see if microck/network-security-setup would activate on your phrasing? Try it before you install — no LLM call, no signup.

Latest benchmark · publisher runVerified

8/21/2026, 04:41 UTC · 23 cases

Lift

+39%

96% vs 57%

Agent turns

0%

1.0 vs 1.0

Tokens

+57%

4,716.39 vs 3,008.61

fail → pass · 10pass → pass · 12pass → fail · 1

Per-case results (23)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-05 Set up network isolation settings for a hybrid Python and Ruby project requiring PyPI (*.pypi.org, pypi.python.org) and RubyGems (rubygems.org). Non-secret configuration includes BUILD_TARGET=production while secret configuration includes DATABASE_PASSWORD. Output the response structured as network configuration, security analysis, and setup instructions.	pass→pass	15,984	10,432	-35%	1	1	0%	2,561	4,643	+81%	0	0	—
case-04 We need sandbox network isolation for a backend service that downloads container images from registry.company.internal:5000 and needs access to api.company.com. We have FEATURE_FLAGS=beta and a PRIVATE_KEY. Return a structured report containing network configuration settings, security analysis, and setup instructions with the target config file path and JSON body.	fail→pass	15,529	9,487	-39%	1	1	0%	2,898	4,517	+56%	0	0	—
case-01 Our enterprise project needs sandbox networking rules configured. We require access to our internal package server at npm.corp.internal, our documentation site docs.corp.com, and outbound proxy routing via http://proxy.corp.com:8080 while keeping standard GitHub access. We also manage LOG_LEVEL=debug and a sensitive DB_PASSWORD. Provide a network configuration plan, a security analysis detailing mitigated risks alongside granted and blocked endpoints, and comprehensive setup instructions featuring the target JSON config file content and validation terminal commands.	fail→pass	19,672	14,461	-26%	1	1	0%	3,884	5,661	+46%	0	0	—
case-02 I am setting up a sandbox environment for an open-source project that relies on standard npm packages, Python PyPI dependencies, and jsDelivr CDN resources. My setup uses NODE_ENV=development and requires an AWS_SECRET_ACCESS_KEY. Please give me the output structured into network configuration details, a security analysis covering threat mitigations and access lists, and setup instructions including config file paths, JSON configuration content, and verification commands.	fail→pass	17,862	10,681	-40%	1	1	0%	3,489	4,718	+35%	0	0	—
case-03 We are working on a highly sensitive banking component that must operate under strict isolation without external registry or domain access. We have environment variables for BUILD_TARGET=staging and a private API_KEY. Generate the complete network configuration, a security analysis outlining access controls and prevented threat vectors, and full setup instructions with the exact settings JSON payload, destination path, and validation commands to verify isolation.	fail→pass	28,477	10,366	-64%	1	1	0%	1,891	4,470	+136%	0	0	—
case-06 Configure network security settings for a developer workstation behind a corporate HTTP proxy located at http://corporate-proxy.company.com:3128. Local requests to localhost and *.internal must bypass the proxy. Provide the output containing network configuration, security analysis, and setup instructions including the configuration JSON content.	fail→pass	16,187	13,990	-14%	1	1	0%	3,207	5,655	+76%	0	0	—
case-07 Configure network isolation for a frontend app that needs access to all subdomains of company-cdn.net and company.cloud, as well as standard Vercel deployments (*.vercel.com). Return network configuration, security analysis, and setup instructions including config location.	pass→pass	19,333	9,519	-51%	1	1	0%	3,371	4,684	+39%	0	0	—
case-08 Create a network isolation configuration for an offline, air-gapped security analysis task. No external HTTP or HTTPS connections should be permitted. Provide a response structured into network configuration, security analysis, and setup instructions.	fail→pass	14,450	8,240	-43%	1	1	0%	2,322	4,089	+76%	0	0	—
case-09 Set up sandbox access for an npm and GitHub development workflow. Include validation commands to test whether package installation works and whether untrusted domains are correctly blocked. Output structured network configuration, security analysis, and setup instructions.	pass→fail	17,476	10,920	-38%	1	1	0%	2,960	4,708	+59%	0	0	—
case-10 We need network security configuration for a microservices team that requires domain matching for api.service.internal as well as deep subdomains like ..company.com. Provide the response with network configuration, security analysis, and setup instructions.	fail→pass	27,067	9,942	-63%	1	1	0%	4,693	4,573	-3%	0	0	—
case-11 Configure enterprise sandbox network settings where HTTP proxy is http://proxy.corp.com:8080 and HTTPS proxy is http://proxy.corp.com:8080, with 127.0.0.1 and localhost in noProxy. Non-sensitive API_BASE_URL is https://api.staging.corp.com. Return network configuration, security analysis, and setup instructions.	pass→pass	13,106	9,280	-29%	1	1	0%	2,456	4,571	+86%	0	0	—
case-12 Provide a network isolation configuration for an automated code reviewer agent to prevent prompt injection attacks from exfiltrating sensitive data to arbitrary attacker-controlled domains. Present your response under network configuration, security analysis, and setup instructions.	pass→pass	15,593	14,301	-8%	1	1	0%	2,592	5,312	+105%	0	0	—
case-13 Configure sandbox network policies for a CI pipeline that needs to pull containers from Docker Hub (registry.hub.docker.com, *.docker.io) and GitHub Container Registry (ghcr.io). Output structured network configuration, security analysis, and setup instructions.	pass→pass	16,313	10,608	-35%	1	1	0%	3,030	4,705	+55%	0	0	—
case-14 Configure sandbox domain whitelist rules for a Java cloud application pulling dependencies from Maven Central (.maven.org) and container images from Azure Container Registry (.azurecr.io). Include LOG_LEVEL=info. Format response with network configuration, security analysis, and setup instructions.	pass→pass	17,148	7,925	-54%	1	1	0%	3,081	4,357	+41%	0	0	—
case-15 We are configuring sandbox environment settings for a project that needs non-secret NODE_ENV=development and requires developers to supply an API_KEY locally without putting raw keys into the sandbox settings file. Return network configuration, security analysis, and setup instructions.	pass→pass	15,890	9,172	-42%	1	1	0%	2,791	4,556	+63%	0	0	—
case-16 Configure trusted network access for a local frontend project running local development servers (localhost:3000). Include validation step to verify local binding. Output network configuration, security analysis, and setup instructions.	fail→pass	19,123	9,830	-49%	1	1	0%	3,515	4,623	+32%	0	0	—
case-17 Set up network isolation policies for a Web application deploying to Netlify (.netlify.com) and connecting to Supabase (.supabase.co). Provide network configuration, security analysis, and setup instructions.	fail→pass	22,353	11,521	-48%	1	1	0%	3,727	4,812	+29%	0	0	—
case-18 Set up network isolation controls to protect against malicious package downloads from unauthorized package indexes. Provide network configuration, security analysis detailing threat mitigations, and setup instructions.	pass→pass	17,725	12,488	-30%	1	1	0%	2,937	5,061	+72%	0	0	—
case-19 Configure custom network sandbox settings for an enterprise team using internal registry npm.company.com alongside standard public package resolution at registry.npmjs.org and source access at *.github.com. Output network configuration, security analysis, and setup instructions.	fail→pass	20,565	8,692	-58%	1	1	0%	3,499	4,343	+24%	0	0	—
case-20 Set up sandbox access for a web application utilizing assets from Unpkg (unpkg.com) and container images from Google Container Registry (gcr.io). Output structured network configuration, security analysis, and setup instructions.	pass→pass	19,558	7,837	-60%	1	1	0%	3,680	4,274	+16%	0	0	—
case-21 Configure a Git pre-commit hook using Gitleaks to scan local commits for exposed AWS keys and database passwords before they are committed to the repository.	pass→pass	12,852	11,872	-8%	1	1	0%	2,449	4,975	+103%	0	0	—
case-22 Write Terraform HCL code to define an AWS VPC Security Group (aws_security_group) that allows inbound SSH on port 22 from CIDR block 10.0.0.0/16 and restricts all outbound traffic.	pass→pass	6,703	8,614	+29%	1	1	0%	1,328	4,338	+227%	0	0	—
case-23 Configure an Nginx web server block for api.company.com to terminate TLS using Let's Encrypt SSL certificates located at /etc/letsencrypt/live/api.company.com/fullchain.pem.	pass→pass	12,892	10,198	-21%	1	1	0%	2,837	4,832	+70%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted. The headline lift of +39 percentage points is the difference between those two pass rates over the 23 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 56.5%
Pass rate with the skill: 95.7%
Cases attempted: 23
Cases counted toward lift: 23
Cases that regressed: 1
Skill version measured: v1
Run completed: 8/21/2026, 04:41 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-04

Prompt

We need sandbox network isolation for a backend service that downloads container images from registry.company.internal:5000 and needs access to api.company.com. We have FEATURE_FLAGS=beta and a PRIVATE_KEY. Return a structured report containing network configuration settings, security analysis, and setup instructions with the target config file path and JSON body.

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signals1 of 4 complete

BenchmarkPass rate WITH skill on a verified run96

Live evalNeeds ≥20 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Efficiency · separate axis (at equal correctness)+0% turns+70% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

Backed by 1 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested 8 days ago

Other measured skills in the registry, with their headline benchmark lift.