How to use the OAuth Security Testing agent skill (offensive-oauth)

What is an agent skill, and what does this one do?

An agent skill is a folder containing a SKILL.md file of instructions that an AI agent loads when a conversation matches the skill's topic. This skill's folder is offensive-oauth, and its file is SKILL.md.

Once installed, when you ask the agent about OAuth, it follows a structured checklist of known weaknesses and checks how a given implementation handles each one. The checklist covers authorization code interception, redirect_uri bypass, CSRF on the OAuth flow, state parameter abuse, open redirector chaining, token leakage via Referer, PKCE bypass, and scope escalation. The skill also explains OAuth best practices and flow mechanics, including OAuth 2.1 and FAPI guidance.

The file is a text methodology, not an automated scanner. The agent reasons over the checklist; it does not run tests against a system for you.

Install the offensive-oauth skill

Download the ZIP from this page and unzip it. You can install the skill in one of two ways:

For a personal setup, unzip it into ~/.claude/skills/ so the offensive-oauth folder sits there. For a single project, unzip it into that project's .claude/skills/ folder.

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

• Personal: ~/.claude/skills/offensive-oauth/
• Project only: .claude/skills/offensive-oauth/
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Trigger the skill with natural requests

You do not need a special command. The skill is meant to activate when the conversation involves OAuth terms such as OAuth 2.0, authorization code, redirect_uri bypass, OAuth CSRF, state parameter, PKCE bypass, scope escalation, token leakage, open redirector, or OAuth attack.

Describe the context in plain language and name the flow or parameter you are interested in. Mention that you are authorized to test the target, because the checklist is written for attacking live targets.

• Ask which redirect_uri checks a login implementation should enforce, and how to test them on an app you are authorized to test.
• Ask whether a missing or reused state parameter is exploitable in a specific OAuth callback.
• Ask for a review of your own app's PKCE handling before release.
• Ask the agent to explain the difference between the Authorization Code and Implicit flows and why Implicit is risky.

Use the checklist on an authorized target

The skill's methodology lists the items to work through. The agent is instructed to follow steps in order unless you say otherwise, consider each technique against your target, track which items are complete, and suggest next steps based on findings.

The hunt list includes intercepting OAuth flows with a proxy such as Burp or ZAP, manipulating redirect_uri parameters, removing or tampering with the state parameter, testing PKCE implementations, inspecting token handling in browsers, checking for client secret leakage, analysing scope handling, testing account linking, reviewing token validation, and examining refresh token security.

• Intercept the flow with Burp or ZAP and note each redirect and parameter.
• Test redirect_uri handling, including path traversal and exact-match behaviour.
• Check whether state is present, unguessable and tied to the session.
• Review PKCE, token validation (aud, azp, iss) and refresh token rotation.

Use it as a developer checklist

If you build a login with Google or GitHub, the skill's OAuth Considerations and OAuth 2.1 Updates sections work well as a review list. They describe PKCE everywhere, exact redirect_uri matching with no wildcards or path traversal, validating aud, azp and iss, ensuring APIs do not accept ID tokens, and refresh token reuse detection.

The mobile and SPA notes add guidance such as using Authorization Code with PKCE in single-page apps, storing tokens in memory, and not shipping client secrets in mobile binaries.

• Confirm PKCE is required by your authorization server and validated on token exchange.
• Confirm redirect_uri matches a pre-registered allowlist exactly.
• Confirm state is unguessable and matched on callback.
• Confirm APIs reject ID tokens and check token audience.

Safety, requirements and limits

The skill lists no requirements. It does not run scripts and does not use the network. It is a set of instructions only, so it cannot scan a target by itself.

The skill is an offensive reference. Use it only on systems you own or have explicit written permission to test. The checklist is written for attacking live targets and could facilitate unauthorized access if misused. It is not a setup guide, so if you only want a simple way to secure a login, pick a guide aimed at that task instead.

• Requires: nothing listed
• Runs scripts: no
• Uses network: no
• Only test systems you own or have explicit written permission to test