How to use the CTF Sandbox Entry agent skill (ctf-sandbox)

What the ctf-sandbox skill makes the agent do

An agent skill is a folder with a SKILL.md file that tells an agent how to handle a certain kind of task. This one is a thin entry gate, not a full exploit toolkit. Its description says it is a PRIMARY for CTF, AWD and 靶场 multi-type orchestration that hands off to a sidecar skill called CTF-Sandbox-Orchestrator.

When triggered, the skill first tells the agent to run a case-initialisation script that sets an authorisation scope and a network profile. It then points the agent to the sidecar orchestrator, which chooses a specific competition sub-skill. Tasks that are clearly pwn, Android APK or IDA reverse engineering are meant to be taken by more specific skills, not this one.

• Sets an authorisation scope and a lab or offline network profile before work starts
• Hands the job to the CTF-Sandbox-Orchestrator sidecar skill
• Leaves pwn, APK and IDA tasks to more specific skills that already match them

Install the ctf-sandbox skill

Download the ZIP from the skill page on this site. To use it with Claude Code, unzip it into ~/.claude/skills/ to make it available everywhere, or into a project's .claude/skills/ folder to limit it to that project.

In the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills. Either way, the skill also depends on files that are not in this ZIP: the referenced ../scripts/case-init.ps1 script, the CTF-Sandbox-Orchestrator skill folder found by relative path, and a routing.json configuration from the wider repository.

• Download the ZIP from this page
• Unzip to ~/.claude/skills/ (all projects) or your project's .claude/skills/ (one project)
• Or upload it in the Claude apps under Settings → Capabilities → Skills
• Check that the required sidecar folder and routing.json are also available

Check the requirements before you start

The skill needs PowerShell to run the case-init script. It also needs the sidecar CTF-Sandbox-Orchestrator skill folder and a routing.json file from the wider repository. The ZIP alone does not contain these, so the skill will not work as a standalone tool.

The skill runs scripts and uses the network, so review what it will do on your machine before you enable it.

• PowerShell, to run ../scripts/case-init.ps1
• The CTF-Sandbox-Orchestrator sidecar skill folder, referenced by relative path
• A routing.json configuration from the wider repository

Trigger the skill with natural requests

The skill activates when your request mentions CTF, AWD, 靶场 or 比赛题 and no more specific pwn, APK or IDA route has already matched. Write the request as you would to a colleague, and say which network setting applies, such as a lab range or offline environment.

Before any external network action, the skill requires the authorisation status to be granted. If you ask for work against a real internet target, the agent should stop until that authorisation exists.

• Example: 'I have a CTF challenge from my class, help me organise the work in a lab range'
• Example: 'We are running an AWD training round on our offline lab network'
• Example: '比赛题 靶场 练习, set up the scope first'
• For clearly pwn or APK tasks, ask for that specific work directly so the more specific skill can take it

Safety notes: scripts, network and authorisation

The first step runs a script, and the skill is designed around scope control. Use the lab or offline network profile for competitions and ranges. The skill's own self-check asks the agent to confirm that it followed case-init and scope first, rather than treating 'the user said CTF' as permission to act on external networks.

The CTF-Sandbox-Orchestrator folder is described as a GPL side package, and its default authorisation is inside the sandbox. The core routing package is MIT and uses a scope.md gate.

• Only run against real external targets after written authorisation is in place
• Use -NetworkProfile lab or offline for competitions and practice ranges
• Do not copy the 40+ competition sub-skills into routing.json; this entry is only a PRIMARY gate
• Review the scripts before running them on your machine

Tips for getting good results

Keep the scope clear in your first message: say whether the target is a lab, an offline range or a training network. Name the challenge type if you know it, so the router can pick the right specialist. If the skill does not trigger, check that the description keywords appear in your request and that the sidecar folder is installed in the expected relative location.

• State the network environment early
• Name the challenge type when you know it (pwn, APK, IDA)
• Confirm the sidecar and routing.json are present if routing fails