How to use the API security testing with Strix skill

What does the api-security-testing skill make the agent do?

APIs differ from web apps. There is no rendered page to crawl, and many serious bugs are authorization problems, such as one user reading another tenant's data, rather than injection problems. The skill points Strix's agents at those specific weaknesses.

The agents enumerate endpoints from an OpenAPI/Swagger file, a Postman collection, a GraphQL schema, or by crawling. They then try to exploit broken object-level authorization (BOLA/IDOR), broken object property level authorization (excessive data exposure and mass assignment), broken function-level authorization, unrestricted resource consumption, SSRF, injection, and auth/token flaws. Every finding is supposed to come with a working proof-of-concept request.

How do you install the skill?

Download the skill ZIP from this page and unzip it. For a personal install, put the folder in ~/.claude/skills/. For a single project, put it in that project's .claude/skills/ folder. In the Claude apps, you can instead upload the ZIP under Settings, then Capabilities, then Skills.

The skill folder contains one file, SKILL.md. The skill itself does not install Strix. You need the Strix CLI working on your machine before the skill is useful. Run strix --version to check. If it fails, use the penetration-testing-with-strix skill to set it up.

• Requirements for a local run: Docker and an LLM API key, or a Strix managed cloud account.
• Without Docker and an LLM key, use the managed path: strix cloud login, then strix cloud scans start, as described in the managed-pentesting-with-strix skill.

What do you need before running a scan?

Collect your inputs first. The agents work much better with a schema and real credentials than with guesswork. Use two test accounts or tokens, ideally in different tenants, so cross-tenant access can be proven. For function-level authorization you also need a low-privilege token and an admin token.

Also provide example object IDs and a list of routes that are out of scope, such as payment or mass-notification endpoints. Only test APIs you own or have written permission to test. The skill sends real attack traffic.

• Schema: OpenAPI/Swagger (.json, .yaml, .yml), a Postman collection, a GraphQL endpoint with introspection, or a gRPC .proto file.
• Two sets of credentials, ideally in different tenants.
• A low-privilege token and an admin token for function-level tests.
• Example object IDs and out-of-scope routes.
• POSTMAN_API_KEY environment variable, only if you pull a Postman collection live.

How do you trigger the skill and run a scan?

Ask your agent in plain language, for example: pentest the staging API at this URL using this OpenAPI file, or audit my API for authorization bugs. The skill is meant to activate on requests to pentest, security-test, audit, or find vulnerabilities in an API, endpoint, or backend service.

The agent will run Strix with the schema passed as a target, not described in prose. The example below uses two test tenants and an admin token. Replace the placeholders with your own values, and keep tokens out of shell history and committed files. Use --instruction-file for long credential blocks.

• Postman export: pass the collection file as a target, or use -t postman://<collection-uuid>.
• Several services: list one target per line in a file and pass --target-list ./targets.txt.
• Backend source: add it as a target for deeper checks on ownership logic.
• gRPC: pass the .proto with --workspace-file. A .proto given as -t fails.
• Supporting notes or wordlists: pass them with --workspace-file ./notes.md.

How do you verify the findings?

Start with strix_runs/<run>/penetration_test_report.md, then read the files in vulnerabilities/. Each vulnerability file contains the exact request that proved the issue. Replay that request yourself, for example with curl, before you report it.

For authorization findings, confirm the response really contains the other tenant's data. An empty 200 response is not proof. findings.sarif uploads to GitHub code scanning, and vulnerabilities.json is a structured index for ticketing.

• Read penetration_test_report.md first.
• Replay each request in vulnerabilities/*.md.
• Check that cross-tenant responses contain real foreign data.

What are the safety limits and tips?

The skill runs scripts and uses the network. The agents send real requests, and they may send harmful ones to endpoints that change money or data. Exclude payment and destructive admin routes explicitly, and prefer a staging environment. Mention rate limits or a WAF so the agents can adapt instead of burning budget on throttled requests.

After fixing issues, use the fix-security-vulnerabilities-with-strix skill, then re-run against the same target to confirm the exploit no longer works. For pull-request checks, use the ci-security-scanning-with-strix skill.

• Use --max-budget to cap spend on a run.
• Fix the authorization check, not just the single endpoint.
• For internal APIs unreachable from your machine, see managed-pentesting-with-strix for the network connector.
• For web apps rather than APIs, see owasp-top-10-testing.