How to use the pentest-agent-os agent skill for pentest planning

What does the pentest-agent-os skill make the agent do?

The skill frames a penetration test as a search problem. Rather than following a preset chain of known exploits, the agent keeps a project knowledge base of verified facts and asks which next action is most useful given what it already knows. Findings are recorded and then verified before they count as confirmed.

The file itself contains only this map and the rules. The detailed techniques live in the other skills it names, such as reconnaissance, web, cloud, Active Directory, binary, blockchain, AI application, wireless and zero-day discovery skills.

• Acts as a skill map: it lists each attack-surface skill and what it covers.
• Sets resident rules for the project knowledge base (pentest-blackboard) and verification (pentest-verification).
• Defines three trigger rules: when identified components should be searched for vulnerabilities, when a request is refused and the agent should switch routes, and when a new credential, domain or IP should be expanded immediately.
• The SKILL.md text is written in Chinese, so expect the agent's working notes and the index to be in Chinese.

What you need before installing

The skill needs an AI agent runtime that can load SKILL.md skills. The index on its own is not enough for the stated workflow. The listing says the detailed methods only become usable when the other skills in the suite are also present, including pentest-blackboard and component-vuln-intel.

The pentest-blackboard component stores project facts in SQLite, so you need that storage working for the workflow described. The listing marks the skill as using the network, and the component-vuln-intel part is described as searching online for vulnerabilities after identifying components.

The repository is listed under the Apache-2.0 licence. The listing does not state a price.

• An agent runtime that loads SKILL.md skills (the listing names Claude's skills folders and the Claude apps as install targets).
• The other suite skills, which were not included in the files reviewed for this listing.
• SQLite-backed project fact storage for pentest-blackboard.
• Network access, since the suite is built to gather intelligence online.
• Written authorisation to test the target.

Install the skill from the ZIP

Download the ZIP from this page and unzip it. For a personal install, place the skill folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder. The skill folder should contain SKILL.md.

If you use the Claude apps instead, upload the skill under Settings, then Capabilities, then Skills.

Because this listing includes only SKILL.md, check that the other suite skills it names are also installed before relying on the detailed methods.

• Personal install: ~/.claude/skills/
• Project install: the project's .claude/skills/ folder
• Claude apps: Settings → Capabilities → Skills, then upload

How do you trigger the skill with a natural request?

The skill description says to use it when starting a full-chain penetration test engagement or when you need the skill map for this suite. Your request should say what you are doing and that you are authorised to do it. The agent then loads this index and decides which attack-surface skill to add next.

For a map only, ask for the structure of the suite. For an engagement, state the target and the scope you have been given. The index tells the agent to load the matching skill for the surface in front of it, for example component identification leads to component-vuln-intel and web work leads to web-attack-methods.

• Map request: 'Explain the skill suite structure and which skill covers web testing.'
• Engagement request: 'I have written authorisation to test the staging app at the scope I describe below. Start a full-chain pentest plan and load the right skill for each surface.'
• Zero-day request: 'Standard searches found nothing for this component. Use the zero-day discovery path.'

Examples of how the skill guides the agent

Suppose the agent identifies a specific software component on a target. The index says to run the component-vuln-intel steps and treat the leads as tentative until verified, then mark them confirmed. If a route is refused, the index says to run the proxy-tool-bootstrap route switch and record a negative result if that fails. Each new credential, domain or IP should be pushed into the project knowledge base straight away.

The index also lists output and conduct skills, including pentest-output-standards for reports and records and redteam-opsec for operational discipline, so the agent has rules for writing up and for staying quiet, not just for finding flaws.

• Identified component: search online for known issues, keep them tentative, verify, then confirm.
• Refused request: switch route, and write a negative fact if no route works.
• New credential or host: expand laterally and record it immediately.

Safety, scope and what to avoid

The listing says the suite is built to attack systems and could cause harm or legal liability if used without written authorisation. It also says it can scan, attempt exploitation and expose data, so it is unsuitable for production systems that cannot tolerate that.

The file is not a learning guide for defence. The content assumes deep offensive-security knowledge. The listing states the file runs no scripts, but the suite uses the network and component-vuln-intel searches online.

• Only test systems you have written authorisation to test.
• Do not point it at production systems that cannot tolerate scanning or exploitation.
• Review what the agent sends over the network before running it.
• Treat the suite as a professional tool, not a beginner course.

Tips for getting useful output

Give the agent the scope, the goal and the starting point in one message. The index is built around a path that is not fixed, so a clear goal lets the agent choose the next action well. Keep the other suite skills installed, because the index only points to them.

Check the project knowledge base for verified facts after each session, since the rules treat unverified leads as tentative.

• State the scope and authorisation at the start.
• Install every named skill, not just SKILL.md.
• Ask for the skill map first if you are new to the suite.