SkillMachine
All skills

Terminology Consistency Auditor

Report

Extracts a termbase from your docs and scores term consistency 0-100 with MQM.

Live output preview

Input Format: Input FormatOutputWatch the Output: Watch the Output

A plan is required to view this content

Choose a plan to access input format, sample outputs, and live previews.

View Plans →

About the skill

What it does

Scans a document set (product/manual/UI/contract text), extracts a concept-oriented termbase, and systematically audits whether the same concept is rendered inconsistently across the set. The core lens is the MQM (Multidimensional Quality Metrics, ASTM F3686) Terminology dimension: it tracks two main violation types — termInconsistency (one concept translated with ≥2 different terms) and termbaseViolation (an approved preferred term not used). Each violation gets an MQM severity weight (Minor=1, Major=5, Critical=10), and a terminology-subset score is computed: Score = clamp(100 − (total weighted penalty / tokens × 1000), 0, 100). The score maps to a PASS (≥95) / CONDITIONAL (85-94) / REWORK (70-84) / FAIL (<70) verdict; any Critical violation (a translated brand/placeholder) triggers a hard-stop capping the verdict at CONDITIONAL.

The termbase is built around concepts, not words (ISO 30042/TBX): morphological variants, synonyms and case variations collapse under a single concept, each assigned a preferred term plus admitted/deprecated variants. Brand names, product codes and placeholders ({0}, %s) are flagged Do-Not-Translate via ITS 2.0 — translating them is a Critical violation.

When to use it

When you suspect the same term is used inconsistently (e.g. "account" → three different renderings); when an inherited TM/termbase needs a term-hygiene measurement; when a pre-delivery terminology QA gate is required; or when a glossary must be built from scratch. Works both monolingually (source-internal variation) and bilingually (source↔target mapping + DNT leakage).

Method / frameworks

  • MQM / ASTM F3686 — Terminology dimension + severity weights (1/5/10); weighted penalty density per 1000 tokens.
  • ISO 30042:2019 (TBX) — concept→langSet→term + termStatus concept-oriented model.
  • ISO 12616 — term↔concept univocity: one concept, one preferred term per target language.
  • ISO 17100:2015 — terminology management as a mandatory revision-step control, here automated.
  • W3C ITS 2.0 — Terminology + Do-Not-Translate tagging.
  • TAUS DQF — MQM-aligned shared error language for LSP workflows (bilingual lens).

How do I use this skill?

You don't "run" a skill — after installing it you just tell the agent your task (e.g. ask for the relevant job), and the skill kicks in by itself when its description matches.

Upload the terminoloji-tutarlilik-denetcisi.zip you downloaded as-is — no packaging needed, the format is already correct (folder at root).

  1. Open Settings → Customize → Skills
  2. Upload → select the terminoloji-tutarlilik-denetcisi.zip you downloaded
  3. Claude reads SKILL.md; the name + description appear. Ready ✅

Scripts run in Anthropic's code-execution environment (sandbox) — not on your machine.