Skip to content

AI skill · Web pentesting · Infrastructure and AD pentesting

AI skill: scoring a finding with CVSS 3.1

Claude Code skill that helps score a finding with CVSS 3.1: it asks for missing facts, justifies each metric, proposes a vector, leaves the score to you.

By Steven Carrier Published

Works with: Claude Code, Claude.ai (projects), agents that support the SKILL.md format

Preview the file
---
name: cvss-scoring
description: "Helps score a security finding with CVSS 3.1: asks for missing facts, proposes each metric with its justification, produces the vector without computing the score and discusses environmental adjustments. Use when the user wants to score, justify or review the severity of a vulnerability."
---

# CVSS 3.1 scoring of a finding

You help a tester score a finding. You propose and justify metrics; you never compute the final score and you run no tests.

## Expected input

The user describes the finding: nature of the weakness, entry point, what was observed (evidence), prerequisites, effect obtained. They may add the client's context (exposure, data criticality).

## Method

1. **Missing facts**: before proposing anything, list the information that determines a metric and that the user has not given. Ask for it as short questions, then wait for the answer. Examples: is access possible from the Internet, from the local network or only locally on the machine? Is an account needed, and at what level? Must a third party (user, administrator) act? Do the vulnerable component and the impacted component fall under the same security authority?
2. **Base metrics**: for each of the eight metrics, propose a value and a one-line justification based on the facts supplied:
   - AV (attack vector): N, A, L, P;
   - AC (attack complexity): L, H;
   - PR (privileges required): N, L, H;
   - UI (user interaction): N, R;
   - S (scope): U, C;
   - C, I, A (confidentiality, integrity, availability): N, L, H.
3. **Uncertainties**: when two values are plausible, give both, say what would tip the balance one way or the other and let the user decide.
4. **Vector**: assemble the vector `CVSS:3.1/AV:…/AC:…/PR:…/UI:…/S:…/C:…/I:…/A:…` from the chosen values.
5. **Calculation**: ask the user to compute the score with the arsenal CVSS calculator (`/en/tools/cvss/` on arsenal.bearops.fr) or with FIRST's official calculator (https://www.first.org/cvss/calculator/3.1), then to give you the result if they want you to include it in a text.
6. **Environment**: if the context is known, discuss the environmental metrics (confidentiality, integrity and availability requirements, compensating controls that modify the base metrics). Present them as points to discuss with the client, not as a final score. Do not invent temporal metrics: they depend on facts only the user knows.

## Output format

1. **Questions** (if facts are missing): numbered list, nothing else.
2. **Metrics table**: metric, proposed value, justification, plausible alternative.
3. **Proposed vector**: one line, no score.
4. **Next step**: compute the score with the calculator indicated.
5. **Environmental adjustments**: short list, only if the context allows it.

## Rules

- Never compute the numeric score or the qualitative severity (Low, Medium, High, Critical) from the vector: point to the calculator. If the user gives you a score obtained from a calculator, use it as is, without recomputing or correcting it; if the vector has changed since that calculation, ask them to recompute.
- Never invent a fact: missing information is a question, not a silent assumption. If you must assume something, write the assumption explicitly.
- Do not inflate the scoring: choose the value the facts establish, not the worst imaginable scenario.
- CVSS 4.0 exists, with other metrics. Work in 3.1 unless asked otherwise; if the user wants 4.0, flag that values do not transpose as they are and point to FIRST's specification (https://www.first.org/cvss/v4.0/specification-document) and calculator (https://www.first.org/cvss/calculator/4.0).
- Remind the user that a base score measures neither the likelihood of real exploitation nor the business risk.
- Provide no exploit payloads.
- Write in the user's language, with no pleasantries or generic conclusion.

Two testers rarely score the same finding in the same way: it all depends on the assumptions made about the attacker, the privileges required or the user interaction. This skill structures that reasoning. It asks the questions that are missing, proposes a value for each metric with its justification, and leaves you to compute the score with a reliable tool.

What the skill produces

  • The list of missing facts, requested before any scoring (network access required, authentication, interaction, scope of the impact).
  • A proposal for each of the eight CVSS 3.1 base metrics, with a one-line justification based on your facts.
  • The full vector (CVSS:3.1/AV:…/AC:…/PR:…/UI:…/S:…/C:…/I:…/A:…), with no score.
  • The points of uncertainty and the plausible alternative values, so that you can decide.
  • A discussion of environmental adjustments (confidentiality, integrity and availability requirements, compensating controls) when the client’s context is known.

The skill never computes the final score: it asks you to compute it with the arsenal CVSS calculator or with FIRST’s official calculator. A calculation done by a language model in its head can be wrong, and a wrong score in a report is hard to retract.

Install the skill

In Claude Code, for all your projects:

mkdir -p ~/.claude/skills/cvss-scoring
curl -fsSL -o ~/.claude/skills/cvss-scoring/SKILL.md \
  https://arsenal.bearops.fr/telechargements/skills/notation-cvss/en/SKILL.md

For a single project, put the file in .claude/skills/cvss-scoring/SKILL.md at the root of the repository. With another agent, add the file’s content to its project instructions.

Use it

Describe the finding and what you actually observed:

Finding: IDOR on GET /api/invoices/{id} of the staging application (target.example).
An authenticated user with no special rights reads other customers' invoices
by changing the identifier. Read-only, no modification observed.
Propose the CVSS 3.1 metrics with justification and the vector. Tell me what is missing.

The model first answers with its questions (is the application exposed to the Internet? can an account be obtained freely?), then with the proposed vector. Paste that vector into the CVSS calculator to get the score and severity, and carry them into the write-up. To format the finding, follow up with the writing a pentest report skill.

CVSS 4.0

Version 4.0 exists, with other metrics (notably attack requirements and the distinction between the vulnerable system and subsequent systems), threat and supplemental metric groups, and a nomenclature stating which groups were used (CVSS-B, CVSS-BT, CVSS-BE, CVSS-BTE). The qualitative severity bands are the same as in 3.1. The skill works in 3.1, which is still the format expected in many contracts and reports. If your client asks for 4.0, say so: the skill flags it and points you to FIRST’s specification and calculator rather than converting the values in its head.

Limitations

  • Scoring remains a judgement: the skill proposes, you decide. Two plausible values for a metric are flagged, not picked at random.
  • A base score measures neither the likelihood of a real attack nor the business risk. Environmental adjustments are discussed, not imposed.
  • The skill does not reproduce the finding and tests nothing: it relies on what you describe.
  • Do not send any client data to an online AI service unless your engagement agreement allows it. Describe the finding without domain names, tokens or personal data where necessary.

Sources

Related resources

  • AI prompt · Web pentesting · Infrastructure and AD pentesting

    Critical vulnerability alert prompt

    Prompt that drafts the urgent client notification when a critical vulnerability is found during testing: facts, impact and immediate containment advice.

    Published

  • AI prompt · Infrastructure and AD pentesting · Web pentesting

    CVE summary prompt

    Prompt that summarises a CVE or vendor advisory: affected versions, conditions, impact, applicability questions for the client, fix and limits.

    Published

  • Tool · Web pentesting · Infrastructure and AD pentesting

    CVSS 3.1 calculator

    Compute the CVSS 3.1 base score, severity and vector string of a vulnerability, or paste an existing vector to review it. Runs in your browser.

    Published

BEAROPS

Need a penetration test?

BEAROPS experts test your applications, infrastructure and cloud, black box or grey box, following OWASP, NIST and ANSSI guidance.