ctf-skills: A Skill Pack That Turns Claude Code, Codex and Gemini CLI into CTF Solvers
Agent skills for solving CTF challenges - web exploitation, binary pwn, crypto, reverse engineering, forensics, OSINT, and more
At a glance
- What is it?
- ljagiello/ctf-skills is an MIT-licensed collection of Agent Skills for web exploitation, pwn, crypto, reverse engineering, forensics and OSINT. It installs with npx and ships a script that pre-installs the toolchains each category needs.
- Who is it for?
- Adopt ctf-skills if you already drive an Agent Skills-capable CLI and want category-level playbooks for web, pwn, crypto, reverse, forensics or OSINT work instead of writing your own prompts per challenge. Skip it if your competition allows no outside assistance, or if you want a standalone solver binary rather than skill files an agent reads.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository received new commits within the last day.
- What is it written in?
- Mainly Python, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.
Star history
- ljagiello/ctf-skills3,717 ★ · +374
| Repository | 2026-09-24 | 2026-11-15 |
|---|---|---|
| ljagiello/ctf-skills | 3.3k | 3.7k |
374 Stars in the last 52 days, measured from the daily GitHub snapshot. Compare with other projects
Editorial analysis
What ctf-skills actually solves for a CTF team
A general-purpose coding agent knows Python and can read a stack trace. It does not know that a challenge involving a truncated MySQL column rewards a specific injection shape, or that a glibc 2.39 build changes how tcache keys behave. ctf-skills packages that category knowledge as Agent Skills: files the agent loads when the task matches, rather than a prompt you paste by hand.
The repository organises this into directories named after challenge classes: ctf-web, ctf-pwn, ctf-crypto, ctf-reverse, ctf-forensics, ctf-osint, ctf-ai-ml, ctf-malware, ctf-misc and ctf-writeup. Each holds a SKILL.md plus supporting technique files. The README's table lists file counts per skill, from 3 files in ctf-ai-ml and ctf-osint up to 22 in ctf-web.
The intended user is someone who already works inside a CLI agent and wants the agent to reach for the right technique without being told. The README states it works with any tool that supports the Agent Skills spec, and names Claude Code as one example. The topics list on the repository also mentions codex, codex-cli, gemini, gemini-cli and opencode.
This is not a solver. Nothing here runs a challenge for you. It is reference material and templates that an agent reads, plus an installer for the underlying tools.
How the skill files and the installer fit together
The mechanism has two halves. The first is the skills themselves: markdown files under each ctf-* directory. The README describes them as covering specific techniques, for instance blind boolean, time-based and error-based SQL injection generators in ctf-web, or a ROP chain template set in ctf-pwn. The ctf-pwn skill ships five runnable pwntools==4.15.0 template scripts in ctf-pwn/scripts/ with yield-based payload generators. That is a concrete artefact you can read before deciding whether the skill is worth loading.
The second half is the environment. scripts/install_ctf_tools.sh is the central entrypoint and takes a mode argument: all, python, pat, apt, brew, gems, go or manual. The README notes the full package lists live in that same script, so the script is the source of truth rather than the README.
One design detail worth noting: pat mode shallow-clones PayloadsAllTheThings into ~/.ctf-tools/PayloadsAllTheThings and links it at ctf-web/payloads/PayloadsAllTheThings. The README explains this keeps bulk payloads out of git, with technique files referencing ctf-web/pat-reference.md and a few exemplar payloads inline. That is a deliberate trade: the repository stays small, but the web skill is less useful until you run pat mode or have network access.
Install logs go to ~/.ctf-tools/, which gives you an audit trail of what the script did.
Installing ctf-skills and running a first challenge
The README gives one install command for the skills themselves:
npx skills add ljagiello/ctf-skillsThat pulls the skill files into your agent's skills location. After it finishes, your Agent Skills-capable CLI should be able to discover the ctf-* skills; the README does not document the exact on-disk path the npx command writes to, so check your tool's skills directory if nothing appears.
Next, set up the toolchain. The README recommends pre-installing before a competition rather than during one:
bash scripts/install_ctf_tools.sh allIf you only care about one category, narrow it. For example, the Python group:
bash scripts/install_ctf_tools.sh pythonBefore committing to a full install, preview it. The README states this skips packages already present:
bash scripts/install_ctf_tools.sh --dry-run allAfter the real run, verify:
bash scripts/install_ctf_tools.sh --verifyIf something failed, --force reinstalls everything regardless of what is already present. Logs land in ~/.ctf-tools/, so a failed package is traceable there.
The README also describes an on-demand path: each skill's SKILL.md has a Prerequisites section listing only the tools that category needs, so you can install as you go when the agent reports a missing tool. For a live event, the pre-install path is the safer one, because a mid-challenge install competes for the same minutes you need for the challenge.
Where ctf-skills stops being the right tool
The repository is a skill pack, not a harness. If your workflow is a single binary that ingests a challenge and emits a flag, this is not that, and the README does not claim otherwise.
Coverage is uneven by design. ctf-web lists 22 files and ctf-crypto 19, while ctf-ai-ml and ctf-osint list 3 each. If your event is mostly OSINT, you are getting a much thinner set of technique files than a web-focused competitor would. The README's tables make this visible, which is at least honest, but it means the pack is not uniform across categories.
The pat dependency is a real failure mode. If the installer cannot reach GitHub to shallow-clone PayloadsAllTheThings, the web skill's payload index is missing, and the README's own description says bulk payloads are deliberately kept out of git. An air-gapped machine needs that clone done earlier.
The installer is also a shell script that touches apt, brew, gems, go and pip. Running it with all on a machine you care about changes system package state broadly. The --dry-run flag exists for exactly this reason and the README documents it, but nothing in the repository suggests a sandboxed or containerised install path. Treat it as a machine you are willing to modify.
Finally, competition rules matter more than tooling. The README does not discuss whether agent assistance is permitted in any specific event. That is your call to make before the event, not after.
ctf-skills versus writing your own agent prompts
The obvious alternative is not another product. It is a private prompt library: your own notes on RSA attacks, your own pwntools templates, kept in a folder and pasted into the agent when needed. That approach costs nothing and fits your exact habits.
The difference is in what gets loaded and when. Agent Skills are discovered by the agent based on the task, so the ctf-crypto file covering, say, Boneh-Durfee or Manger attacks is available without you remembering it exists at 3am. A prompt library only helps if you recall the right note.
The second difference is the installer. A personal notes folder has no equivalent of scripts/install_ctf_tools.sh, so tool setup stays manual. The ctf-skills script centralises apt, brew, gems, go, pip and manual installs behind one command with a dry-run mode and a verify mode.
The trade is editorial control. A skill file written by someone else may push a technique you disagree with, or may be thinner than your own notes on a topic you know well. You can read the files before adopting them; they are markdown in a public repository, and the ctf-pwn templates are runnable code you can inspect directly.
Maintenance, licence and upgrade cost
The repository is MIT licensed, which permits commercial and private use with the licence and copyright notice retained. That is the standard permissive arrangement; the repository does not state any additional terms, and nothing here should be read as legal advice.
The last push was on 2026-09-13, and the repository is not archived. The repository has no published GitHub releases, so there is no versioned upgrade path to follow. Upgrades are whatever lands on main.
That has a practical consequence: the installer script is also the package list, per the README, so a pull that changes scripts/install_ctf_tools.sh changes what gets installed on your machine. Re-running with --dry-run all after a pull shows the delta before you apply it.
The upgrade cost is otherwise low. Skill files are markdown, so a pull is cheap and there is nothing to compile. The expensive part is re-running the installer, which touches system package managers. If you have a working pre-competition image, pulling new skill files without re-running the installer is the low-risk middle path.
Editorial conclusion
Adopt ctf-skills if you already drive an Agent Skills-capable CLI and want category-level playbooks for web, pwn, crypto, reverse, forensics or OSINT work instead of writing your own prompts per challenge. Skip it if your competition allows no outside assistance, or if you want a standalone solver binary rather than skill files an agent reads. Before a real event, run bash scripts/install_ctf_tools.sh --dry-run all to see the package list, then --verify after the real run, and check the last push date on the repository, which was 2026-09-13, before assuming a given technique file is current.
Frequently asked questions
What are the CTF skills taught by Claude in ctf-skills?
The repository groups them by challenge class: ctf-web, ctf-pwn, ctf-crypto, ctf-reverse, ctf-forensics, ctf-osint, ctf-ai-ml, ctf-malware, ctf-misc and ctf-writeup. The README's table lists the techniques per skill, such as SQL injection generators under ctf-web and ROP chain templates under ctf-pwn. Claude Code is named in the README as one tool that supports the Agent Skills spec.
Does ctf-skills work with Codex, Gemini CLI and opencode?
The README states it works with any tool that supports the Agent Skills spec and names Claude Code as an example. The repository's topics list also includes codex, codex-cli, gemini, gemini-cli and opencode. The README does not give per-tool setup instructions beyond the npx install command.
What skills do I need for CTF, and does ctf-skills cover them?
The repository covers web exploitation, binary pwn, crypto, reverse engineering, forensics, OSINT, AI/ML, malware and misc, plus a writeup directory. Coverage is uneven: ctf-web lists 22 files and ctf-crypto 19, while ctf-ai-ml and ctf-osint list 3 each. Which categories matter depends on the event you are entering.
What tools does the ctf-skills installer need?
The installer modes are python, pat, apt, brew, gems, go and manual, so it expects whichever of those package managers your platform provides. The full package lists live in scripts/install_ctf_tools.sh rather than the README. Install logs are saved to ~/.ctf-tools/.
Can I preview what ctf-skills would install before running it?
Yes. The README documents bash scripts/install_ctf_tools.sh --dry-run all, which skips packages already present, and bash scripts/install_ctf_tools.sh --verify to check what is already installed. The --force flag reinstalls everything regardless of what is present.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/ljagiello-ctf-skills)Star history chart for your README
[](https://hysenlabs.com/projects/ljagiello-ctf-skills)Was this useful?