How to use the Proxy Routing and Tool Bootstrap agent skill

What an agent skill is and what this one makes the agent do

An agent skill is a folder with a SKILL.md file that gives an AI agent instructions for a specific task. The agent loads the skill when your request matches its description. This skill's description says it applies when a test is blocked by 403, 429, a WAF or a timeout, when tools are missing, or when out-of-band confirmation is needed.

When triggered, the skill tells the agent not to simply retry a blocked request. It runs a fixed sequence: look up the target's country, fetch SOCKS5 proxies for that country from ProxyScrape, test each proxy against the target, fall back to HTTP proxies, and finally try a local Tor client on port 9050. It also covers writing Python substitutes for missing tools, generating wordlists from the target's name and keywords, and using interactsh-client or a VPS listener to see callbacks.

Install the skill: download the ZIP and unzip it

Download the ZIP of the skill from this page. Unzip it into one of two places. For a personal setup, use ~/.claude/skills/. For a single project, use the .claude/skills/ folder inside that project. The skill folder should contain SKILL.md.

If you use the Claude apps instead, open Settings, go to Capabilities, then Skills, and upload the ZIP there.

• Personal: ~/.claude/skills/ (unzip so SKILL.md sits in proxy-tool-bootstrap/)
• Project: .claude/skills/ inside your project folder
• Claude apps: Settings → Capabilities → Skills → upload the ZIP

Check the requirements before you start

The skill depends on other things being present. It does not bundle its own tools. The source lists curl and python3 with requests, bs4, hashlib, pycryptodome and scapy (scapy is optional, used only for substitutes). It needs network access to ip-api.com and api.proxyscrape.com. The Tor fallback needs a local Tor client listening on port 9050. Out-of-band confirmation needs interactsh-client, a VPS, or ngrok or cloudflared.

The skill also references the companion pentest-blackboard skill, which the agent uses for its findings log. Without it, the trigger rule the proxy skill cites will not resolve.

• curl
• python3 with requests, bs4, hashlib, pycryptodome (scapy optional)
• Access to ip-api.com and api.proxyscrape.com
• Tor client on 127.0.0.1:9050 (Tor fallback only)
• interactsh-client, a VPS, or ngrok/cloudflared (OOB)
• The pentest-blackboard skill

Trigger the skill with natural requests

You do not need special syntax. Describe the situation in plain language and name the target you are authorized to test. The agent matches your request against the skill description.

Good trigger phrasings follow the blocked-or-missing pattern: say what is failing and what you want next. Include the scope so the agent stays on the in-scope target.

• "The login endpoint on the in-scope host returns 429 after a few requests. Switch routes and retry."
• "Nmap isn't installed. Write a Python port check for the top ports on 203.0.113.10 and report open ones." (replace with your own in-scope address)
• "Confirm whether the parameter in the in-scope form triggers blind callbacks using interactsh."
• "Build a wordlist for the company name in the scope document and use it for directory brute-forcing."

How the proxy fallback sequence works

The fallback runs in order, and each step runs only if the previous one fails. The agent first resolves the target's country with ip-api.com, then requests SOCKS5 proxies for that country code from ProxyScrape, and tests each one with curl against the target URL. A 200 response means the proxy works, and the agent replays the blocked request through it. Otherwise it moves to the next proxy.

The source says SOCKS5 is preferred for probing. HTTP proxies can insert their own error pages (such as 502), which makes it hard to tell the target's real response from the proxy's. HTTP proxies are only suggested for anonymity or rotation against a target already confirmed reachable. Rotation guidance in the source: switch immediately on 429 or 403, rotate every 20 requests, and for Cloudflare use a proxy pool with random 2 to 5 second delays.

• Step 1: look up the target's country (ip-api.com)
• Step 2: fetch SOCKS5 proxies for that country (ProxyScrape)
• Step 3: test each proxy against the target with curl --socks5
• Step 4: on 200, reuse the proxy for the blocked request; otherwise try the next
• Step 5: if all SOCKS5 fail, fetch and test HTTP proxies
• Step 6: if all proxies fail, try Tor at 127.0.0.1:9050

Use tool bootstrapping and generated wordlists

When a tool is missing, the skill tells the agent to check with which and write a Python replacement. The source gives these substitutes: a socket-based port scan instead of nmap, requests-based directory brute-forcing instead of ffuf, a manual payload check instead of sqlmap, requests-based brute force instead of hydra, and requests-based known-payload checks instead of nuclei. Helpers for crawling (requests and bs4), encoding, hashing, encryption and sniffing (scapy) are also listed.

For wordlists, the skill builds variants from the target domain or company name, pulls keywords from web pages, combines usernames with years and special characters, and includes default vendor credentials.

• No nmap: socket port scan
• No ffuf: requests directory brute force
• No sqlmap: manual payload detection
• No hydra: requests brute force
• No nuclei: requests with known payloads

Confirm blind vulnerabilities with out-of-band callbacks

Blind vulnerabilities give no visible response, so the skill uses out-of-band (OOB) infrastructure. Options in the source are interactsh-client, which provides an oast.fun domain, or a VPS running a simple listener, or an ngrok or cloudflared tunnel. The skill treats a received DNS query or HTTP request as confirmation and only then records a finding.

Keep the findings log in the pentest-blackboard skill so each confirmed item has its evidence.

• interactsh-client gives you an oast.fun domain
• VPS: python3 -m http.server or nc to watch for connections
• ngrok or cloudflared for a tunnel
• Confirmed = a DNS query or HTTP request arrives

Safety notes: scripts and network use

The skill runs scripts and uses the network. It sends your target's IP to ip-api.com and requests proxy lists from api.proxyscrape.com. Your test traffic then passes through proxies that come from public, unvetted sources, so the operators of those proxies may be able to log or intercept it. Do not use this skill on sensitive or regulated environments.

The skill is built to get past blocks. Use it only on systems you own or have written permission to test, and only within the agreed scope. Read the scripts the agent generates before running them.

• Data sent to ip-api.com and ProxyScrape includes the target address and country lookups
• Public proxies may log or intercept traffic
• Only test systems you own or have written permission to test
• Review generated scripts before execution