How to use the Offensive Crypto Attacks agent skill

What an agent skill is and what this one does

An agent skill is a folder containing a SKILL.md file with instructions that an AI agent loads when a task matches the skill's description. This skill contains only a SKILL.md file.

The skill gives the agent a six-step workflow: identify cryptographic touchpoints such as cookies, tokens, API parameters and signed URLs; fingerprint the algorithm and mode; classify the vulnerability class; select a tool or custom script; execute the attack; and document the weakness, its root cause and the remediation path.

It covers padding oracle attacks against CBC-mode ciphers with PKCS7 padding, ECB exploitation (block cut-and-paste and byte-at-a-time decryption), hash length extension against SHA1, SHA256 and MD5-based MACs, RSA weaknesses (small public exponent, common modulus, Bleichenbacher PKCS1v1.5 padding oracle, Coppersmith's method), weak PRNG and MT19937 state recovery, timing side-channels, AES-GCM nonce reuse, and weak key derivation. It maps these to CWE-327, CWE-328 and CWE-330.

Install the skill

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

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

• Personal: ~/.claude/skills/offensive-crypto-attacks/
• Project: .claude/skills/offensive-crypto-attacks/ inside your project

Install the required tools

The skill needs Python 3, PyCryptodome, the requests library, padbuster, RsaCtfTool and HashPump. The source gives the hashpumpy package name for Python bindings, so install that with pip. The source does not give install commands for the other tools, so follow each project's own documentation.

• Python 3
• PyCryptodome (used for building custom exploit payloads)
• requests library
• padbuster
• RsaCtfTool
• HashPump (hashpumpy Python bindings)

Trigger the skill with natural requests

Describe the target and the goal in plain language and name the cryptographic object you want analysed. The agent matches the request to the skill's description. Include the scope of your authorization so the agent works within it.

Good requests name a concrete artifact, for example an encrypted session cookie in an application you are authorized to test, and ask for a specific outcome such as decrypting the value or checking whether it can be forged.

• Analyse the encrypted cookie in this authorized test app and tell me whether the padding leaks information.
• Check whether this API's signed token uses a length-extendable hash like SHA256 over secret plus message.
• This is a CTF challenge: the RSA public exponent is 3 and I have two ciphertexts. Identify the weakness.
• Review how our app derives keys from the user password and report missing salts or low iteration counts.

Example: a padding oracle workflow

The skill tells the agent to first find an oracle by submitting modified ciphertext and watching for differences in HTTP status codes, error messages, timing or application behaviour. Once an oracle is confirmed, padbuster can automate decryption. The source's example passes the target URL, the encrypted value, a block size (8 or 16), and flags for cookies, encoding and an error string. Replace the placeholder values with your own authorized target.

If padbuster cannot handle the target's encoding or transport, the skill includes a custom Python example that uses requests and PyCryptodome to decrypt one block via Vaudenay's attack. Its oracle check returns True when the server's response is not a 500, so you must adapt that signal to your target.

What the skill needs and what it does on your machine

According to the skill's listing, it does not run scripts on its own and it does use the network. Its attacks work by sending many crafted requests to the target, so the network use is the central safety consideration. The skill's own example code is shown as text for the agent to adapt; review any script before you let the agent run it.

• Runs scripts: no
• Uses network: yes
• Requires: Python 3, PyCryptodome, requests, padbuster, RsaCtfTool, HashPump

Safety notes before you run anything

Use the skill only against systems you own or have written permission to test. The techniques are illegal to use against third parties. For production-critical targets, get a rules-of-engagement agreement first, because the scripts send many crafted requests that can disrupt services.

The skill's own emphasis is black-box identification before targeted exploitation. Ask the agent to confirm the oracle and scope before it sends bulk traffic.

• Confirm written authorization and scope.
• Use a test or CTF environment when learning.
• Agree rules of engagement before testing production systems.
• Ask the agent to document root cause and remediation at the end.

Tips for better results

Give the agent the full context: the encoding (Base64, hex), the block size if known, and what a valid versus invalid response looks like. The skill's workflow starts with fingerprinting, so these details save time.

Ask for the root cause and fix, not just the exploit. The skill asks the agent to document the weakness and remediation, which is useful for developers auditing their own code.

• Provide sample requests and responses from the target.
• State the expected error signal for the padding oracle.
• Ask for the CWE mapping and remediation in the report.