How to use the API Fuzzing for Bug Bounty agent skill

What the api-fuzzing-bug-bounty skill makes the agent do

An agent skill is a folder with a SKILL.md file that gives an AI agent extra instructions for a specific task. This skill covers API reconnaissance, authentication testing, IDOR (insecure direct object reference) testing, injection testing, HTTP method testing, GraphQL-specific probes and endpoint bypass checks.

The skill is a set of techniques, not an automated scanner. The source describes its payloads as hypotheses that need human judgment to confirm.

• Find API documentation and hidden endpoints, such as Swagger or OpenAPI files
• Check login paths and whether auth endpoints are rate limited
• Test whether one user can access another user's object IDs
• Probe for SQL, command and XXE injection, SSRF and path traversal patterns
• Run GraphQL introspection, schema discovery and nested-query checks

Install the skill from a ZIP file

Download the ZIP from this page and unzip it. The skill folder contains SKILL.md. Place the folder in ~/.claude/skills/ to make it available to you across projects, or in a project's .claude/skills/ folder to limit it to that project.

You can also upload the skill in the Claude apps under Settings → Capabilities → Skills.

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

Trigger the skill with natural requests

You do not need a special command. Describe the task in plain language and name the target and your authorization. The agent will then load the skill when the request matches its description.

Good requests state the goal, the target and the scope. For example, you can ask it to plan API reconnaissance for an in-scope staging host, or to review a login flow for rate limiting. Always include the exact target, because the skill asks for it before running any probing command.

• Plan an authorized API test of my staging API at the URL I give you, and list the read-only steps first
• Review this GraphQL schema for IDOR risks in the user query using my test accounts
• Explain how to check whether the login endpoint on my lab app is rate limited

Understand the safety gates before you run anything

The skill contains a mandatory confirmation gate. Before any command that probes, exploits, changes, persists on, extracts data from or attempts credential access against a target, the agent must ask you for the exact target, ask you to confirm written authorization and permitted scope, show the exact commands with their expected effect, and wait for explicit confirmation in the conversation.

Without that confirmation the agent stays read-only and gives defensive guidance. The skill also says to prefer a sandbox, disposable VM or controlled lab. The skill instructs you to scope rate, concurrency, duration, write or delete methods, SSRF destinations and callback endpoints in your rules of engagement, and to stop on rate limits or unexpected access to real personal data.

• Use test accounts and read-only canary resources
• DoS-style deep recursion, credential attacks and destructive method tests need separate explicit approval
• Never evade rate limits through IP rotation

What the skill needs and what it does on your machine

The skill's files list only SKILL.md. The source does not include runnable scripts. It references tools you must obtain yourself: Burp Suite or a similar intercepting proxy (optional), Python 3 for helper scripts such as json2paths.py (referenced but not included), Kiterunner, GraphCrawler, graphw00f, clairvoyance and SecLists wordlists.

The skill uses the network because the techniques send requests to the target you name. It does not run scripts itself.

• Runs scripts: no
• Uses network: yes
• Licence: MIT

Example: a read-only API reconnaissance pass

A typical first request is to look for API documentation on a host you are authorized to test. The skill lists common documentation paths to check, such as /swagger.json, /openapi.json and /api-docs. These are plain GET requests to see what is published.

Only after you confirm the target and scope would the agent suggest running a discovery tool such as Kiterunner, which the source shows as kr scan with a routes wordlist.

Tips for getting useful results

Give the agent your test accounts and a canary resource you own, so IDOR checks compare your own data against another test account rather than real users. Ask it to explain expected effects before each command. Treat every finding as a hypothesis until you have confirmed it manually and documented it.

Keep the mobile and web APIs separate in your notes, since the skill advises not to assume they share the same security controls.

• Start with defensive review and read-only checks before any probing
• Record the scope, timestamps and every command you approved
• Use a lab or disposable VM for injection and DoS-style tests