domain-language
Audit established domain terms across glossaries, ADRs, code, schemas, APIs, tests, UI, and external contracts.
Skill by hhellbusch
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/hhellbusch-domain-language
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: documents, design, security
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/domain-language
Tools it pre-approves 4
From allowed-tools in SKILL.md: the agent may use these without asking you each time.
ReadGrepGlobShell
What's inside
- Skill name
domain-language- Description
- Audit established domain terms across glossaries, ADRs, code, schemas, APIs, tests, UI, and external contracts. Report which source owns each meaning. Use when the user says /domain-language, "term audit", "glossary drift", "who owns this word", or "do these names mean the same thing?".
- License
- MIT
- Allowed tools
- Read, Grep, Glob, Shell
- Instructions
- 108 lines in SKILL.md · view on GitHub
Files (1)
- SKILL.md
SKILL.md (first lines)
# Domain language — term alignment (invoked) <objective> Report how an **established** domain term is used across surfaces, and which source owns each meaning. Evidence only. Do not invent a new canonical meaning. Do not rewrite an ADR, rename code, or “fix” history in the same pass. If a human decision is required, classify it contradiction or unresolved and stop. </objective> <constraints> - Established terms only — not greenfield naming brainstorms - Quote or cite the surface (path, symbol, field, sentence). No paraphrase-as-proof - One term (or a tight cluster the user named) per invoke - Classify each mismatch; do not collapse them into “inconsistent” - Not `/shoshin` (is the problem framed right?), `/craft` (engineering tradeoff), or an ADR (record a decision) </constraints> <process> ### 1. Name the term From `$ARGUMENTS`, or ask once which term. If a path is given, extract the term and still search the other surfaces. ### 2. Collect surfaces Search what exists in the repo (skip empty categories): | Surface | Look for | |---------|----------| | Glossary / docs | definitions, READMEs that define the word | | ADR | decision text that binds a meaning | | Code | types, functions, comments that define behavior | | Schema | fields, enums, DB names | | API | paths, payload fields, error codes | | Tests | fixtures and assertions that encode meaning | | UI | labels and copy | | External contract | OpenAPI, public docs, partner payloads | Note the **owner** when a surface says it is authoritative (ADR status Accepted, published contract, glossary titled as such). If none claims ownership, say **unowned**. ### 3. Classify each difference One class per difference. Precedence when two seem to fit: 1. **Contradiction** — same token, incompatible meanings (lifecycle vs identity is the usual case), even if both sides claim to be the owner 2. **Intentional translation** — a documented boundary (glossary, ADR, contract) says both names are the same thing at different edges 3. **Legacy name** — old token remains, behavior follows the current owner, and no document explains the boundary 4. **Unresolved** — sources differ, you cannot show they are incompatible, and no accepted owner exists Do not emit contradiction until the surfaces that exist have been checked. A missing search is not a contradiction. ### 4. Report Use the output format. Name a human decision only for **contradiction** or **unresolved**. Do not pick the winner. </process> <output_format> ``` ## Domain language — [term] | Surface | Meaning (cite) | Owner? | |---------|----------------|--------| | … | path: quote | yes / no / unowned | ### Differences - **[class]:** [one sentence + cites] ### Do not change yet [What this audit refuses to rewrite] ### Needs a decision? yes (who should decide) | no ``` </output_format> <classification_key> Author check for the classes. Do not execute these against the user's repo, and do not append them to a real audit. **Intentional translation, not legacy.** Public field `accountId`. Column `user_id`. Glossary: “accountId is the customer-facing id; user_id is the storage key for the same row.” The word “legacy” in that sentence does not make the class legacy name — the boundary is documented. Do not rename the column. **Legacy name.** Code still says `user_id`. Docs and API say `accountId` for the same row. Nothing documents a boundary. Behavior follows `accountId`. Class: **legacy name**. **Contradiction.** Glossary: `retired` = product no longer sold. Identity service: `retired` = credential revoked, person record still active. Tests assert both. Class: **contradiction**. Report both. Do not pick one. </classification_key> <failure_modes> - Silent canon: choosing a meaning and editing code/docs in the same turn - Single-surface audit: only reading the glossary - Calling every alias a contradiction - Rewriting an Accepted ADR because a new name “seems clearer” </failure_modes> <success_criteria> - Every meaning has a cite - Each difference has one class from the table - No renames or new definitions committed as part of the audit - The classification key cases classify as specified, and are not pasted into the user's audit </success_criteria>
From hhellbusch/zanshin-pi-extension, MIT.
More from hhellbusch/zanshin-pi-extension
- 0.565score
shoshin
hhellbusch
Surfaces load-bearing assumptions against source artifacts before building.
SkillAFreshness: AClaude Code +9 - 0.565score
yomi
hhellbusch
Second-order lens (読み — reading ahead): and then what?
SkillAFreshness: AClaude Code +9 - 0.565score
craft
hhellbusch
Apply engineering principles to code or design — DRY, KISS, SRP, YAGNI, convention over configuration, orchestration vs program, shared change scope, phased…
SkillAFreshness: AClaude Code +9 - 0.565score
kihon
hhellbusch
Basics / fixed forms (基本): shell, secrets, vault, sql, git, k8s, testing, lint, ansible, helm, kustomize, python, typescript, structured edit.
SkillAFreshness: AClaude Code +9 - 0.565score
unslop
hhellbusch
Cut AI tells from writing.
SkillAFreshness: AClaude Code +9 - 0.565score
kaeshi
hhellbusch
Inversion lens (返し — reversal): what would guarantee failure?
SkillAFreshness: AClaude Code +9 - 0.559score
lab-journal
hhellbusch
Append one lab-journal entry after a meaningful surprise, or skip.
SkillAFreshness: AClaude Code +9 - 0.558score
spar
hhellbusch
Adversarial review of any content — essays, designs, plans, or ideas
SkillAFreshness: AClaude Code - 0.558score
whats-next
hhellbusch
Analyze the current conversation and create a handoff document for continuing this work in a fresh context
SkillAFreshness: AClaude Code - 0.558score
checkpoint
hhellbusch
Save current session state mid-session — faster than /whats-next, designed for crash recovery
SkillAFreshness: AClaude Code +9
Signals
- Stars
- 0
- Forks
- 0
- Open issues
- 0
- Last commit (this folder)
- Oct 5, 2026 (24 days ago)
- Last release
- Oct 5, 2026 (24 days ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from hhellbusch
- 0.565score
shoshin
hhellbusch
Surfaces load-bearing assumptions against source artifacts before building.
SkillAFreshness: AClaude Code +9 - 0.565score
yomi
hhellbusch
Second-order lens (読み — reading ahead): and then what?
SkillAFreshness: AClaude Code +9 - 0.565score
craft
hhellbusch
Apply engineering principles to code or design — DRY, KISS, SRP, YAGNI, convention over configuration, orchestration vs program, shared change scope, phased…
SkillAFreshness: AClaude Code +9 - 0.565score
kihon
hhellbusch
Basics / fixed forms (基本): shell, secrets, vault, sql, git, k8s, testing, lint, ansible, helm, kustomize, python, typescript, structured edit.
SkillAFreshness: AClaude Code +9 - 0.565score
unslop
hhellbusch
Cut AI tells from writing.
SkillAFreshness: AClaude Code +9 - 0.565score
kaeshi
hhellbusch
Inversion lens (返し — reversal): what would guarantee failure?
SkillAFreshness: AClaude Code +9
Cite this page
domain-language skill. Agent Mart. https://agentmart.dev/skills/hhellbusch-domain-language (last updated 2026-10-24)