Chicio Labs logo
> CHICIO LABS_Pixels. Code. Unplugged.

Glossary Browser

A Claude Code plugin that keeps a repository's ubiquitous language in view while you work. It reads the glossary the repository already has (GLOSSARY-MAP.md, each context's GLOSSARY.md, the ADRs), shows it in a band above the prompt and a pane you can explore, and its Term Check steers you and the model away from the words the glossary says to avoid.

It is a mod: a plugin of function hooks that runs inside Claude Code (terminal or desktop).

Install

/plugin marketplace add chicio/chicio-labs
/plugin install glossary-browser@chicio-labs
/reload-plugins

Works with Matt Pocock's domain-modeling skills

The Glossary Browser reads exactly the files that Matt Pocock's domain-modeling skill writes, so it pairs with that skill and with grill-with-docs, the grilling session that runs domain-modeling and records the terms and decisions as they settle. Grill a plan, and the terms the session resolves show up in the band and the pane, and the Term Check starts steering toward them. Install the skills with:

npx skills add mattpocock/skills -s domain-modeling grill-with-docs grilling

It follows the skill's current convention, GLOSSARY.md and GLOSSARY-MAP.md (GLOSSARY-FORMAT.md). Versions of the skill before 2026-08-15 wrote CONTEXT.md and CONTEXT-MAP.md: rename them to adopt it.

What your repository needs

Either a single GLOSSARY.md at the repository root (one context), or a GLOSSARY-MAP.md there that links each context's glossary:

# Glossary Map

## Contexts

- [Ordering](./src/ordering/GLOSSARY.md): receives and tracks customer orders
- [Billing](./src/billing/GLOSSARY.md): generates invoices

## Relationships

- **Ordering → Billing**: Ordering emits OrderPlaced; Billing invoices it

Each GLOSSARY.md lists its terms under ## Language, optionally grouped by ### sections:

**Invoice**:
A request for payment sent to a customer after delivery.
_Avoid_: bill, payment request

ADRs are read from docs/adr/*.md at the root (system-wide) and inside each context's folder. Without a glossary the plugin stays quiet.

Using it

  • The band above the prompt names the contexts and counts the terms and ADRs. Focus it (ctrl+x tab, or click) and press g to open the pane. When the Term Check has flagged words, they appear in the band: press 1 to 3 to open a flagged term, x to clear the flags.
  • /glossary [term] opens the pane, on that term when one is given.
  • The pane lists the contexts on the left, each expandable to its terms (by section) and its ADRs, plus the system-wide ADRs and the map's Relationships. The right side shows the selection: a term's definition, its Avoid entry, the words the Term Check flags for it, and where the same word means something in another context; an ADR rendered as Markdown. Type in the filter to search every term, definition and ADR title. Esc closes it.

The glossary is read live, and re-read when a GLOSSARY.md, the map or an ADR is edited in the session. The plugin never writes it.

The Term Check

It looks for Avoid words in two places:

  • Your prompts. The prompt goes through exactly as typed, with a note only the model reads: each word, its context and the canonical term ('task' is an Avoid word in Agentic Delivery: when it means Work Unit, say Work Unit). The model uses the canonical terms in its reply and its work, and ignores a line when the word was meant in another sense. Every context's list applies, since a prompt has no path.
  • The model's edits to Markdown (.md, .mdx). An edit to a file inside a context's folder is checked against that context's list. If it uses an Avoid word, the edit is refused with the canonical terms, so the model rewrites it; when the word is meant (a quote, code, another sense), sending the same edit again unchanged lets it through. An edit to a file outside every context is checked against all of them and only flagged; with a single root GLOSSARY.md, its one context owns every file. GLOSSARY.md and GLOSSARY-MAP.md themselves are never checked.

What it does not count as a hit: code blocks and inline code, HTML or JSX tags, link targets, an Avoid word that is part of a canonical name (the glossary's "design" inside "Design System"), and Avoid entries that are guidance rather than words ("using it for a block inside a page").

Configuration

Option Default Effect
denyEdits true Refuse Markdown edits that use the owning context's Avoid words. Off, they are only flagged.

Set it with claude plugin configure glossary-browser@chicio-labs.

Limits

The Term Check is a convenience, not enforcement: it runs only in Claude Code, and the retry lets any word through by design. A rule every contributor and every tool must follow belongs in lint or CI.

Development

From the repository root:

claude plugin validate claude-plugins/glossary-browser
claude plugin test claude-plugins/glossary-browser

CI runs both on every push. tsc -p claude-plugins/glossary-browser type-checks the mod once a Claude Code session has loaded it, since the engine writes the type declarations it extends at load time. Releases go through the release-plugin.yml workflow, which bumps version in .claude-plugin/plugin.json, writes CHANGELOG.md and tags glossary-browser--v<version>.

Project info

Glossary Browser
>_
Type:Claude Code plugin
#
Version:0.2.0
/
Workspace:claude-plugins/glossary-browser