How to use the Web App Penetration Testing agent skill

What the skill makes the agent do

The skill directs Strix to run a black-box penetration test against a live URL, a staging environment, or a local dev server. It looks for real vulnerabilities such as auth bypass, broken access control, IDOR, injection, XSS, SSRF, and business logic flaws.

Each finding is validated with a working proof-of-concept rather than a signature match, so the report shows an exploit that was actually run. The skill can also use source code when you provide it, which improves coverage of authorization and business logic.

Install the skill

Download the skill ZIP from this page and unzip it. For a personal install, place the folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder.

In the Claude apps you can instead upload the skill under Settings → Capabilities → Skills.

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

Set up what the skill needs

The skill runs on the Strix CLI (usestrix/strix), which is typically installed via Docker. You also need an LLM API key or a managed app.strix.ai cloud account. If you test a localhost dev server, install Docker Desktop.

Test account credentials for the target app are recommended. Most real vulnerabilities sit behind login, and without an account the agents only see public pages.

• Strix CLI installed, typically via Docker
• An LLM API key or an app.strix.ai cloud account
• Docker Desktop if testing a localhost dev server
• Test account credentials for the target app (recommended)

Trigger the skill with a natural request

You do not need to name the skill. Ask for a pentest, security test, hack, or audit of your web app, website, web application, or staging site, and the agent will pick it up. Include the target URL and anything the agent needs to log in.

Before any scan starts, the agent should confirm that the target is yours or that you are authorized to test it, which environment it is, which paths are out of scope, and what credentials it can use. Answer these questions in your request so the agent does not have to guess.

• Pentest my staging site at https://staging.example.com. Test account is qa@example.com. Stay out of billing.
• Security-test our web application at http://host.docker.internal:3000 (my local dev server).
• Audit the access control between the two seeded orgs on our staging app.

Run a scan from the command line

The skill's example runs Strix non-interactively against a staging URL with a budget cap. The --instruction flag carries the test account, the scope, and what to focus on.

Use --instruction-file instead when the instructions are long. Include how to log in if the flow is unusual, such as a magic link, SSO, or an MFA-exempt test user. Always set --max-budget.

• Give the agent two accounts when possible, so it can try cross-account access.
• Add your repository with a second -t flag for deeper white-box coverage.
• Use http://host.docker.internal:3000 to reach a dev server on your host from the sandbox.
• Scan modes: quick for a fast dev-loop pass, standard (about 30 minutes) for a normal review, deep for pre-release checks.

Read the results and check the exit code

Start with strix_runs/<run>/penetration_test_report.md, then open the per-finding files in vulnerabilities/. Each one contains its proof-of-concept, so re-run it yourself before you report the issue.

The exit code tells you the outcome: 0 means no validated vulnerabilities in what was analyzed, 2 means vulnerabilities were found, and 1 means a fatal error. A 0 is not proof of full coverage. If the budget or turn cap was hit, the scan wraps up early, so check the status and cost in run.json.

• 0: no validated vulnerabilities in what was analyzed
• 2: vulnerabilities found
• 1: fatal error

Fix the issues and re-test

Patch the root cause of each finding, then run Strix against the same target again to confirm the exploit no longer works. Re-testing is the only reliable check that a fix landed. The source names companion skills for this step and for CI: fix-security-vulnerabilities-with-strix and ci-security-scanning-with-strix.