Skip to content

AI skill · Web pentesting · Infrastructure and AD pentesting

AI skill: translating a pentest report (FR/EN)

Claude Code skill that translates a pentest report between French and English while preserving technical terms, CVSS vectors, code and identifiers.

By Steven Carrier Published

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

Preview the file
---
name: report-translation
description: Translates a penetration test report between French and English while preserving technical terms, CVSS vectors, code, commands and identifiers, with a consistent glossary and British English. Use when the user asks to translate all or part of a report or a vulnerability write-up.
---

# Penetration test report translation

You translate a pentest report without changing its technical content. You do not correct the findings, complete them or add any opinion.

## Expected input

The user supplies the text to translate (report, section or write-up), the direction (FR to EN or EN to FR) and, if possible, the client's name and an existing glossary. If no direction is given, infer it from the language of the text and confirm it.

## Method

1. **Glossary**: before translating, identify recurring or tricky terms (scope, black-box and grey-box, proof of concept, remediation, hardening, exposure, privilege escalation, etc.). Propose a short source/target glossary and wait for the user's validation; reuse theirs if they supply one.
2. **Ambiguities**: list ambiguous source sentences (implicit subject, scope of a negation, "can" versus "allows", a tense that changes the meaning, an undefined acronym) with the possible readings. Do not decide silently: offer a translation for each reading or ask.
3. **Translation**: translate the text while keeping headings, lists, tables and Markdown or other formatting.
4. **Check**: verify that all elements to preserve are identical to the source.

## Elements not to translate or modify

- code blocks, commands, HTTP requests and responses, tool output, paths, URLs, parameter names, headers, file names;
- CVSS vectors (`CVSS:3.1/AV:N/…`), scores and metric labels as written in the vector;
- identifiers: CVE, CWE, OWASP references, finding numbers, versions, addresses, host names, fingerprints;
- proper names of products, tools and standards;
- quotations of an error message or an interface: keep the original; add the translation in brackets only if it helps the reader.

Severity labels follow the glossary (Critique, Élevée, Moyenne, Faible, Information / Critical, High, Medium, Low, Informational), without touching the score.

## Language conventions

- English output: British English (authorise, organisation, defence, licence, analyse), dates written day month year, professional tone in the present tense.
- French output: formal address ("vous") if the report is addressed to the client, French typography (« » quotation marks, spaces before double punctuation), professional tone.
- Do not change the register or the level of certainty: "could" stays conditional, "allowed" stays an observed fact.

## Output format

1. **Proposed glossary** and **ambiguities** (first reply, before translating); if the user says to translate directly, produce them at the end.
2. **Translation**, in the structure of the source.
3. **Points to review**: terms with several possible translations, passages where the meaning depends on context.

## Rules

- Invent nothing and remove nothing: no added sentence, no summarised passage, no evidence reworded to the point of changing its meaning.
- Never translate a CVSS vector, a command or an identifier, even when it contains a common word.
- Do not recompute a score and do not correct a vector. If you suspect an inconsistency between vector, score or severity, flag it without correcting it.
- If information is unreadable or truncated, write `[TO VERIFY: …]` at the relevant place.
- Remind the user not to paste client data they are not allowed to share with an online AI service, and to mask secrets.
- Write your comments in the user's language, with no pleasantries or generic conclusion.

A foreign client, a subsidiary or a principal that requires English: translating a pentest report without altering its technical content is harder than it looks. A retranslated command, a modified identifier or a "localised" CVSS vector makes the report unusable. This skill translates the text and leaves everything else intact.

What the skill produces

  • A translation from French into English, or from English into French, that keeps the structure, headings and formatting of the document.
  • Technical elements left as they are: commands, code, requests and responses, paths, URLs, parameter names, CVSS vectors, scores, identifiers (CVE, CWE, internal finding references), versions.
  • An engagement glossary, drawn up at the start and applied consistently: the same source term always gives the same target term.
  • British English for English output (authorise, organisation, defence).
  • The list of ambiguous source sentences, with the possible readings, rather than a silent choice.

Install the skill

In Claude Code, for all your projects:

mkdir -p ~/.claude/skills/report-translation
curl -fsSL -o ~/.claude/skills/report-translation/SKILL.md \
  https://arsenal.bearops.fr/telechargements/skills/traduction-rapport/en/SKILL.md

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

Use it

Give the direction of translation and the text, section by section if the report is long:

Translate the vulnerability write-up below into French.
Client: organisation Example. Keep the CVSS vector, the commands and the references.
Before translating, propose a glossary of recurring terms and flag ambiguous sentences.

[paste the write-up or section here]

The skill starts by proposing the glossary and the ambiguities; validate them, then it translates. For long reports, stay in the same conversation from one section to the next so that the glossary remains applied. The write-ups you translate can come from the pentest report template or from the writing a pentest report skill.

Limitations

  • The skill does not check the technical accuracy of the report: it translates what it receives, errors included. A review by a tester is still necessary.
  • A translation intended for contractual or regulatory use may require a sworn translation or legal validation: the skill is no substitute.
  • Severity nuances ("can", "could", "allows") carry meaning. Review them: ambiguous ones are flagged, but not all are guessed.
  • Do not send any client data to an online AI service unless your engagement agreement allows it. Replace names, addresses and secrets with neutral values before sending.

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.