How to use the Application Security Testing agent skill

What the skill makes the agent do

An agent skill is a folder of instructions that an AI agent loads when a request matches it. This skill is the entry point for requests like "make my application secure" where the target is not yet a single URL or repository. Its job is to decide which part of your product needs which kind of test, point to the companion skill that runs each test, and merge the results into one ranked plan.

Findings are ordered by proven impact rather than scanner severity. Autonomous agents try to exploit each issue and show whether it works, so the plan reflects what an attacker could actually reach. The skill also requires the agent to state what was not tested and to check whether any run stopped early because of a budget or turn limit.

Install the skill

You can install the skill in one of two ways. Download the ZIP from the skill page, then either unzip it into ~/.claude/skills/ to make it available across your projects, or unzip it into a project's .claude/skills/ folder to make it available only to that project.

Alternatively, upload the ZIP in the Claude apps under Settings → Capabilities → Skills.

• Personal install: unzip into ~/.claude/skills/
• Project install: unzip into the project's .claude/skills/ folder
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Set up Strix before your first run

This skill does not install Strix itself. Installation, LLM setup, and the full list of command-line flags live in the penetration-testing-with-strix companion skill, which you should read first if the strix --version command fails.

Strix is a command-line tool that typically runs with Docker. You also need an LLM API key, or a Strix managed cloud account. If you have no Docker and no LLM key, the managed path is available through strix cloud login and strix cloud scans start, described in the managed-pentesting-with-strix skill.

• Check that the tool runs with strix --version
• Confirm Docker is available if you use the local path
• Have an LLM API key or a Strix managed cloud account

Trigger the skill with natural requests

You do not need special syntax. Describe what you want in plain language and the agent will load the skill when your request matches its description. The skill is designed for requests such as an application security review or audit, an appsec assessment, vulnerability scanning across your stack, a security review before a launch, a customer security questionnaire, or a request where you do not yet know which kind of security test you need.

Before the agent scans anything, it will ask you to map your assets. Have these answers ready: your source repositories and languages, whether a staging deployment exists, your APIs and any schema files, whether you can create two test accounts in different tenants, and any out-of-scope paths or budget and time limits.

• "Run an application security review of our product before the launch next month."
• "We have a customer security questionnaire. Help us get evidence-based findings for our app."
• "We have a monorepo, a staging site and a public API. Work out which security tests we need."
• "Run an appsec assessment and give me a ranked remediation plan."

Follow the test sequence

For a first assessment, the skill recommends a sequence. Start by reviewing the code, since it is the cheapest run and maps the authorization model. Next, pentest a staging environment with credentials and pass the repository as a second target so the agents keep source context. Finally, add CI scanning so later regressions are caught without another manual pass.

Run one asset at a time and read each report before starting the next. Findings from the code review make the live run more focused. The skill maps each asset type to its companion skill: code to find-security-vulnerabilities-in-code, live web apps to web-app-penetration-testing, APIs to api-security-testing, OWASP-mapped assessments to owasp-top-10-testing, and pull requests to ci-security-scanning-with-strix.

• Step 1: code review
• Step 2: staging pentest with credentials and the repo as a second target
• Step 3: CI scanning for regressions

Read the results and build one plan

Findings are written per run into strix_runs/<run>/. The skill tells the agent to merge them into one list and rank them. Validated exploits reachable without authentication come first, then validated cross-tenant or privilege-escalation issues, then validated issues that need an authenticated account. Unproven observations such as configuration, dependency, and hardening notes come last and are labelled as unproven rather than presented as confirmed vulnerabilities.

Duplicates are common, because the same root cause often appears in both the code review and the live pentest. The agent should merge these. Check each run's run.json status and cost against your --max-budget limit. An empty result from a run that was cut short does not mean the system is clean.

• Validated, unauthenticated exploits first
• Validated cross-tenant or privilege-escalation issues next
• Validated authenticated issues after that
• Unproven notes last, clearly labelled

Be clear about what was not tested, then fix

A good assessment states what was not covered. The skill names likely gaps: assets with no staging environment, categories a black-box run cannot reach such as logging and alerting, supply-chain integrity, and insecure design, and any run that hit its budget or turn cap.

When you are ready to fix issues, the skill hands off to fix-security-vulnerabilities-with-strix, which re-runs Strix against each fix to check that the exploit no longer works.