How to use the Cloud Penetration Testing agent skill

What is this skill and what does it make the agent do?

The skill outlines methodologies for performing authorized penetration testing against AWS, Azure, and GCP cloud environments. Once installed, the agent follows a structured workflow: define scope under the shared responsibility model, enumerate the attack surface, test IAM privilege escalation, test for SSRF to cloud metadata services, examine lateral movement, and then test persistence and detection.

The skill names cloud-specific tools such as Pacu, ScoutSuite, Prowler and CloudFox, and frames findings against the MITRE ATT&CK Cloud matrix. It is a methodology with example commands, so the agent writes and suggests commands for you to review and run yourself.

• Scope definition under the shared responsibility model, with a sample scope document
• Reconnaissance with ScoutSuite, Prowler, CloudFox and AWS CLI commands
• IAM privilege escalation checks using Pacu
• SSRF testing against IMDS at 169.254.169.254 for AWS, Azure and GCP
• Lateral movement via cross-account role trusts and shared resource access
• Persistence and detection testing, with a cleanup step

How do you install the skill?

Download the ZIP for this skill from this page and unzip it. For a personal install, place the unzipped skill folder in ~/.claude/skills/. To make it available only inside one project, place it in that project's .claude/skills/ folder.

If you use the Claude apps instead, upload the skill under Settings → Capabilities → Skills.

• Personal install: ~/.claude/skills/
• Project install: .claude/skills/ inside your project
• Claude apps: Settings → Capabilities → Skills, then upload

What do you need before you start?

The skill assumes you already have written authorization from the account owner and a signed scope document. Nothing in the workflow replaces that paperwork. The skill's own prerequisites also call for an isolated test account or a production account with explicitly scoped breakglass procedures.

The skill's scripts and workflow reference the following. Install the ones you plan to use.

• Python 3 with the AWS CLI installed and configured (used by scripts/agent.py)
• AWS CLI, Pacu, ScoutSuite, Prowler, CloudFox and curl
• Valid AWS credentials for the authorized target account
• Written authorization from the account owner

How do you trigger the skill with natural requests?

You do not need special syntax. Describe the authorized job in plain language and name the account or cloud. The agent should use the skill's methodology when your request matches its scope.

Include the context that the skill asks for: that you are authorized, what is in scope, and which account profile you are using. Without that, the agent may ask questions before proceeding.

• "I'm authorized to test the AWS account behind profile target-account. Walk me through an IAM privilege escalation review and give me the commands to review."
• "Help me write a scope document for a cloud penetration test of our production AWS account, following the shared responsibility model."
• "Check whether our EC2 instances still allow IMDSv1, which would make them vulnerable to SSRF credential theft."
• "Map our cross-account role trusts and suggest how to test lateral movement, then classify findings against MITRE ATT&CK Cloud."

What does a typical example look like?

A common starting point is reconnaissance. The skill includes commands that list public S3 buckets and find EC2 instances whose metadata options show HttpTokens. The agent can explain each command and show the output format you should expect, then you decide what to run.

The skill also provides a Python script, scripts/agent.py, and a reference file, references/api-reference.md. According to that reference, the script is run with python agent.py --profile target-account --output pentest_report.json and enumerates IAM users, roles, cross-account trusts, IMDSv1 instances, public S3 buckets, and Lambda secrets.

What safety limits should you know about?

Several workflow steps change the state of an account. The skill includes commands that create IAM users and access keys, attach AdministratorAccess, and stop CloudTrail logging, all labelled as authorized-test-only. Run them only in accounts where you have written permission, and always follow the cleanup step to remove test artifacts.

The script and the commands use the network and run against live cloud APIs. The skill excludes testing cloud provider infrastructure itself and DDoS or volumetric testing. Treat the output of any command as sensitive, because enumeration can expose credentials and secrets.

• Use only accounts you have written authorization to test
• Do not test cloud provider infrastructure or run DDoS or volumetric tests
• Review every command before running it; the agent suggests, you execute
• Commands in the persistence step modify IAM and logging; clean up afterwards
• Protect output such as Lambda environment variables, which can contain secrets

Tips for getting better results

Give the agent a precise scope and a named profile so it does not guess at targets. Ask it to explain why each command is useful, and to map each finding to a MITRE ATT&CK Cloud technique and a remediation. Ask for a report structure that separates in-scope results from anything you should not have touched.

• Paste your signed scope document so the agent can keep to its boundaries
• Ask for one step at a time and check results before moving on
• Ask for findings classified under the MITRE ATT&CK Cloud matrix
• Use an isolated test account when practising techniques