Guide · Agent skill
How to use the Capability Primitive Search agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
Capability Primitive Search is an agent skill, a folder of instructions that an AI agent loads when a task matches it. This one is a penetration-testing playbook, written mostly in Chinese with English keywords, that teaches the agent to treat each weakness as a building block rather than a single critical bug. This guide explains what the skill changes in the agent's behaviour, how to install it, how to trigger it, and the safety limits you should respect before using it.
Capability Primitive SearchAIPentest/CyberStrikeAI · Apache-2.0 Download the skill Quick start
- 1Download and unzip. Download the skill ZIP from this page and unzip it so that SKILL.md sits in a capability-primitive-search folder.
- 2Choose where to install. Copy the folder to ~/.claude/skills/ for personal use, or to a project's .claude/skills/ folder for one project.
- 3Or upload in the Claude apps. In the Claude apps, open Settings → Capabilities → Skills and upload the skill there.
- 4Restart the session. Reopen or restart your agent session so the new skill is loaded.
- 5Confirm scope. Write down the authorized target, the allowed actions and what is out of bounds before you start.
- 6Make a natural request. Describe your findings and your goal in plain language, without naming the skill.
- 7Verify each step. Test every link in a proposed chain yourself and record the evidence before treating it as real.
What the skill makes the agent do
The skill's core idea is that remote code execution is not one bug but a set of capabilities that combine. Instead of asking whether a target has an RCE vulnerability, the agent records what it can do, such as read a file, write a file, execute a command, or make a server-side request (SSRF), along with the limits on each.
The skill lists six equations that each reach code execution from different combinations of primitives. It also maps low-severity findings to primitives, for example an information leak that exposes a key, or a file upload that can be chained with path traversal.
When the obvious chain is missing, the skill tells the agent to search forward from its current capabilities and backward from a goal such as command execution. It also points to legitimate admin features, credential reuse across services, and cross-domain reuse of capabilities as places to look.
- Forward search: ask what each capability unlocks and what pairs of capabilities combine into.
- Backward search: pick a goal, choose the equation closest to the current situation, and treat missing pieces as sub-goals.
- Feature-as-primitive: admin features such as task schedulers, plugins, file managers or SQL consoles may already provide execution or read/write power.
- Creative mode: when known combinations run out, re-examine capability limits and look for alternative sinks.
Install the skill from the ZIP
Download the ZIP for the skill from this page and unzip it. The skill is a single file named SKILL.md, so the unzipped folder should contain that file.
For a personal install, place the folder in ~/.claude/skills/. For a single project, place it in the project's .claude/skills/ folder instead. Restart or reopen your agent session so it picks up the new skill.
If you use the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.
- Personal install: ~/.claude/skills/
- Project install: the project's .claude/skills/ folder
- Claude apps: Settings → Capabilities → Skills
~/.claude/skills/capability-primitive-search/SKILL.mdTrigger the skill with natural requests
The skill's description says it applies when there is no single RCE, when low-severity vulnerabilities need chaining, or when a novel attack chain must be derived. You do not need to name the skill. Describe the situation in plain terms and state the goal and scope.
Good requests name the authorized target, what you already found, and what you want to reach. Stating the scope matters, because the skill itself does not check whether you have permission to test a system.
- "In my authorised test of this staging app, I have an info leak exposing a .git folder and an SSRF that only sends GET requests. Help me find a chain to command execution."
- "I found no critical bug on this host. Using the capabilities I already have, search forward and backward for a path to code execution."
- "Our red team has credentials for one service. Suggest where else they might be reused within the agreed scope."
Safety notes before you run it
This skill is offensive. It has no authorization checks and gives no guidance on scope. Use it only against systems you own or have written permission to test. The skill's own guidance is to verify each step of a chain before treating it as confirmed, which is also the safest way to work.
The skill does not list any required dependencies, does not run scripts, and does not use the network on its own. Any network activity comes from the agent and tools you run it with, so control those tools and keep to the agreed scope. Avoid pointing it at production systems unless the scope explicitly allows it, since attack steps can cause unintended damage.
- Only test systems you own or have written permission to test.
- Keep to the agreed scope and keep records of each verified step.
- Do not run it against live production systems without a safe scope.
- The skill is licensed under Apache-2.0 according to the repository listing.
What people use it for
- Finding a multi-step path in an authorized red-team exercise when no single critical bug exists.
- Teaching students how low-severity weaknesses such as info leaks or SSRF can combine into serious compromise.
- Generating candidate attack chains for a bug bounty target that has only medium or low findings, then testing them within program rules.
- Checking whether credentials found in one service are reused in another within the agreed scope.
Frequently asked questions
How do I install the capability-primitive-search skill?
Download the ZIP from this page and unzip it. Place the capability-primitive-search folder in ~/.claude/skills/ for personal use or in a project's .claude/skills/ folder, or upload it under Settings → Capabilities → Skills in the Claude apps.
Is the capability-primitive-search skill free?
The skill is listed under the Apache-2.0 licence in its repository. Check the licence file in the source repository for the terms that apply to your use.
What does the skill need to run?
The listing states no required dependencies, no scripts and no network use. It is a single SKILL.md instruction file that guides the agent's reasoning.
Is this skill legal to use?
Use it only against systems you own or have written permission to test. The skill is offensive and does not check authorization, so the legal responsibility sits with you.
Why does the skill write in Chinese?
The playbook is written in Chinese with English keywords such as read, write, exec and ssrf. Agents can usually follow the English terms, but you may want to ask the agent to summarize the skill in your own language.
What are alternatives to this skill?
The source does not compare it with other tools, so this guide cannot name specific alternatives. If you need general security testing methods, look for other penetration-testing skills or established testing guides and compare their scope and safety notes.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
This skill suits security professionals on authorized penetration tests or red-team work, students learning how weak findings combine, and bug bounty hunters who need chain ideas. It is a poor fit for anyone testing systems without written permission, for general coding or writing help, or for live production environments without a safe scope.
Download the skill