How to use the Pentest Output Standards agent skill

What the skill makes the agent do

The skill is a set of output rules, not a tool that scans anything. Once it is loaded, the agent writes its analysis and reasoning in Chinese, while tool commands, code and CVE IDs stay in English.

Each step is recorded with a fixed structure: the current mode (single agent, deep, plan_execute or supervisor), the blackboard points it read, its decision reasoning, the action, a result summary, and whether it saved a fact or vulnerability record.

Verified findings are written up with a name, severity, affected versions, exploit conditions, exploit steps (request and response), evidence, impact and fix recommendations. Failed tests are logged as negative results with the method, outcome and reason for exclusion, so they are not repeated. Every write-type change to a target is logged in a ledger with a rollback command.

Install the skill (ZIP download or Claude apps upload)

Download the ZIP for this skill from the directory page and unzip it. The skill folder contains SKILL.md, the only file listed for this skill.

For a personal install, unzip it into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder. Alternatively, in the Claude apps, open Settings → Capabilities → Skills and upload the skill there.

• Personal install location: ~/.claude/skills/
• Project install location: .claude/skills/ inside your project
• Claude apps: Settings → Capabilities → Skills, then upload

Trigger the skill with natural requests

The skill's description says to use it when reporting findings, maintaining a change ledger, or formatting pentest output. You do not need a special command; ask for that kind of work in plain language and the agent can pick the skill up from its description.

Name the task clearly. Mention the finding, the change you made, or the negative result you want recorded, so the agent knows which rule set applies.

• Write up the verified SQL injection on the login endpoint in the standard finding format.
• Log the webshell I uploaded to the staging host in the change ledger with a rollback command.
• Record the directory brute-force and SMB checks as negative results so we don't repeat them.
• Give me the overview of current progress and next steps from the project facts.

Example: what a finished finding looks like

When a finding is verified, the output follows the template: name, severity, affected versions, exploit conditions, steps with request and response, evidence marked as actual output, impact and fix. The rules also say to save the finding to the host platform's vulnerability record and to store the reproduction context as separate project facts.

The ledger uses one row per write-type change with columns for number, time, host, type, location, content and rollback command. Changes to configuration files require a backup first, and the backup path goes into the ledger. Ledger types named in the rules include webshell, new account, configuration change, persistence, permission change, credential request, database change and network change.

What you need before you start

The skill only works inside a host agent environment that provides the tools it references, such as the project-fact tools (upsert_project_fact, search_project_facts, list_project_facts) and the vulnerability recorder (record_vulnerability). The source names CyberStrikeAI as an example host. Without those tools, the skill's rules still describe the format, but the save steps cannot run.

The skill runs no scripts and uses no network access, according to its listing. Its only file is SKILL.md.

• A host agent environment that provides the project-fact and vulnerability-recording tools
• Written authorization for any system you test

Safety notes

The skill is built for offensive security work, but it has no authorization check of its own. It does not verify that you are permitted to test a target. Confirm written authorization before any test, and use it only on systems you are allowed to assess.

The skill's rules say the agent must not clean up changes on its own. Ledger entries and rollback scripts are delivered to you, and you decide what to revert.

• No built-in authorization check: you are responsible for scope.
• The skill itself does not run scripts or access the network.
• Review rollback commands before running them against a target.

Tips for better results

Keep the host tools connected, because the skill's save steps depend on them. Ask for the change ledger at the end of a job so the full list and the reverse-order rollback script are produced together.

If several rounds of testing produce only negative results with no new attack surface, the skill's dead-end rule tells the agent to re-review existing project facts, change domain or depth, or ask you for clarification or authorization boundaries. Expect that behaviour and answer those questions.

• Ask for negative results explicitly so dead ends are recorded.
• Request the progress overview after each round to keep the next step clear.