glossary（glossary 搜索）

Glossary — Ubiquitous Language Manager

Manage a project's shared vocabulary as version-controlled YAML files. The glossary bridges the gap between what developers call things in code (cart_item), what users see in the UI ("Shopping Bag"), and what the team says in meetings ("basket"). It's the single source of truth for terminology — consumable by humans, AI tools, CI pipelines, and future web frontends alike.

Inspired by Domain-Driven Design's Ubiquitous Language and compatible with Contextive.

Glossary location and discovery

The glossary can live anywhere in the project. Users may choose a custom path (e.g. docs/terminology/, content/glossary/). Always resolve the glossary root before reading or writing.

Discovery order:

1. Explicit pointer — Check project config for a stated glossary path:

- CLAUDE.md — section like "Glossary: path/to/glossary" or "Glossary lives in …" - .cursorrules or AGENTS.md — terminology or glossary path - README.md — project structure or "Glossary" section

1. Conventional locations (if no pointer found):

- glossary/glossary.config.yml (project root) - docs/glossary/glossary.config.yml - .claude/glossary/glossary.config.yml

1. Search — Look for any glossary.config.yml in the project. If exactly one is found, use its directory as the glossary root. If multiple, list them and ask the user which to use.

If no glossary exists: Proceed to Scaffolding and create the glossary at the project root (glossary/) unless the user asks for a different location. If they specify a path, create it there and add a pointer in CLAUDE.md or.cursorrules so future discovery finds it.

All paths in this skill (e.g. "read the context file", "append to the correct glossary file") are relative to the resolved glossary root (the directory containing glossary.config.yml).

Modes

The skill operates in three modes based on what the user asks.

Explore (default)

When the user wants to look something up. Resolve the glossary root (see Glossary location and discovery), then read the relevant glossary files under that root and answer the question.

• "What does Order mean?" — look up the term, show its definition, context, code_name, ui_label, and relationships. If it appears in multiple contexts, show both definitions and explain how they differ.
• "What terms are in checkout?" — read the checkout context file, list all terms with brief definitions.
• "How does Cart relate to Checkout Session?" — trace the relationship chain through related fields and explain the path. Include any when conditions.
• "Show me stale terms" — scan all glossary files for terms with status: deprecated or status: proposed.
• "What do we call X in the frontend?" — look up the term's ui_label. Or reverse: given a UI label, find the code_name.

For the full term schema and field reference, read references/term-schema.md.

Add

When the user wants to add or update terms. Walk through it conversationally — the user shouldn't need to write YAML.

Step 1: Find the glossary. Resolve the glossary root using Glossary location and discovery above. If no glossary exists, switch to Scaffolding (see below).

Step 2: Understand the term. Ask the user:

• What's the term?
• What does it mean? (in plain language)
• What do developers call it in code? (code_name)
• What do users see in the UI? (ui_label)
• Which context does it belong to?

Don't ask all questions at once — infer what you can from context. If the user says "add Cart to the checkout glossary," you already know the name and context. Just ask for the definition and the fields you can't infer.

Step 3: Check for conflicts. Before adding:

• Does this term already exist in this context? → Offer to update instead
• Does it exist in a different context? → Flag as cross-context term, ask if intentional
• Is it an alias of an existing term? → Suggest adding to that term's aliases instead
• Does the code_name or ui_label conflict with another term? → Flag the collision

Step 4: Suggest relationships. Based on existing terms in the glossary:

• "I see Cart already exists. Does Cart contain this new term?"
• "Checkout Session is in the same context. How does this term relate to it?"

Only suggest relationships that make sense. Don't force the user to define relationships for every term — orphan terms are fine, especially early on.

Step 5: Write the YAML. Generate the term entry with the separator comment, append it to the correct context file under the resolved glossary root, and regenerate relationships.md in that root if any relationships were defined.

Show the user the generated YAML before writing it. Let them adjust.

Maintain

When the user wants to check the health of the glossary. Resolve the glossary root (see Glossary location and discovery), then load all glossary files under that root and run checks:

• Orphan detection — terms with no relationships (no related and not referenced by any other term). List them. Not necessarily a problem, but worth reviewing.
• Stale term review — terms with status: deprecated. Ask if they should be removed or if the replacement term needs updating.
• Consistency check — scan the codebase for identifiers that match or resemble code_name values. Report:

- Code identifiers not in the glossary (potential missing terms) - Glossary code_name values that don't appear in the code (potentially stale)

• Cross-context conflict report — terms with the same name in multiple context files. Show how their definitions differ. Ask if the divergence is intentional.
• Relationship integrity — check that every term referenced in a related field actually exists in the glossary. Flag broken references.
• Relationship diagram — regenerate relationships.md with Mermaid diagrams.

Present findings as a report with suggested actions. Don't auto-fix anything — let the user decide.

Scaffolding

When the skill is invoked in a project with no glossary, bootstrap the structure.

Step 1: Ask about the project. "What's this project about? What are the main domain areas?"

Step 2: Propose bounded contexts. Based on the answer, suggest 2-4 initial contexts. Always include core for terms shared across the whole project. Present as a checklist:

I'd suggest starting with these contexts: - core — terms shared across the whole project (User, Permission, Notification) - auth — authentication and authorization - billing — payments, subscriptions, invoicing Want to adjust these?

Step 3: Scaffold the directory. Create the glossary directory at the project root as glossary/ (unless the user asked for a different path). That directory must contain:

• glossary.config.yml with project name, description, and context entries
• One .glossary.yml file per context, each with the context header and empty terms: []
• An empty relationships.md

Step 4: Seed from the codebase. Scan for likely domain terms:

• Model/entity names from source code (class names, type definitions)
• Component names from frontend code
• Terms from existing .feature files
• Domain language from README and docs

Present findings as suggestions — never add terms automatically:

I found these terms in your codebase that aren't in the glossary yet: - User (appears in 34 files) - Session (appears in 12 files) - CartItem (appears in 8 files — likely "Cart Item") Want to add definitions for any of these?

Step 5: Update project configuration. If the project has any of these files, add a glossary pointer so discovery finds the glossary next time:

• CLAUDE.md — add a Glossary section with the path (e.g. Glossary: glossary/ or the custom path used)
• .cursorrules — add a terminology note with the glossary path
• README.md — add a mention in the project structure section

If you scaffolded at a custom path (user requested), always add or update a pointer. Only touch files that already exist. Don't create new config files for tools the project doesn't use.

How Other Skills Use the Glossary

The glossary is plain YAML; its root is resolved the same way as in this skill. Any skill or tool can read it — no special API needed. The file format is the API.

When another skill (e.g. gherkin) needs terminology:

1. Resolve the glossary root using the same Glossary location and discovery (pointer → conventional locations → search).
2. Read glossary.config.yml in that root to find context files.
3. Read the relevant context file(s) based on scope (paths relative to the glossary root).
4. Use terms in its output (ui_label for user-facing text, code_name for technical text).

The glossary skill doesn't need to be invoked for this — other skills read the files directly once the root is known.

Integration examples

Loading Strategy

Don't load the entire glossary into context. Use tiered loading:

Use the scope field in glossary.config.yml to determine which context files are relevant. If working in src/checkout/, load checkout.glossary.yml + core.glossary.yml from the glossary root. If scope is ambiguous, ask: "Which context are you working in?"

Compatibility

The YAML format is compatible with Contextive. Contextive's VS Code extension can read these files and show hover definitions in the editor — it picks up name and definition, ignoring extra fields like code_name and related.