CTFs (Capture The Flag competitions) are hacking competitions where players race to exploit deliberately vulnerable systems and recover hidden flags. Challenges cover everything from web applications and compiled binaries to cryptography and digital forensics.
ctf-skills teaches AI coding agents like Claude Code and Codex to approach these challenges the way an experienced CTF player would: what to look at first, which tools to use, and which techniques are likely to pay off.
The project is a collection of skills, each focused on a different type of challenge, built using the open Agent Skills format. This means any compatible agent can use them. A coordinating skill analyzes each new challenge, identifies its type, and delegates it to the appropriate specialist. Other skills help the agent find and learn from existing writeups when it gets stuck, then document its own solution once it finds the flag.
The part I care about most is measuring whether these skills actually make a difference. The repository includes an evaluation suite that tests three things: whether the agent selects the right skill for a challenge, whether it follows the rules defined by a skill even when prompted to cut corners, and whether having the skills improves its success rate on small, offline challenges compared to working without them.
The project is open source under the MIT license, and I maintain it in my spare time. If it saves your team hours during a competition or security assessment, consider sponsoring it on GitHub.