Glossary

• Reference document -- look up any term or tool.

Tools

• Claude Code — An AI coding agent from Anthropic that runs in your terminal or VS Code. It reads your codebase, writes code, runs commands, and creates git commits. It is not a chat bot — it is an agent that can take actions on your computer.

• **CLAUDE.md** — A markdown file at your project root that acts as Claude Code's persistent memory for the project. It is loaded automatically at the start of every session. Think of it as the project briefing document that Claude reads before starting work.

• Beads (bd) — A git-backed issue tracker designed for AI coding agents, created by Steve Yegge. It stores issues as structured data in your git repo. Claude Code reads Beads at session start (via hooks) so it always knows what needs to be done. The CLI command is bd. The bd ready command loads the current active issue context into a Claude Code session, giving Claude the task details, acceptance criteria, and any comments without manual copy-pasting.

• Context7 — An MCP server by Upstash that provides up-to-date library and framework documentation. When Claude needs to look up a React API or a database library, Context7 fetches the current docs instead of Claude relying on its training data (which may be outdated). It is NOT a plugin — it connects via the MCP protocol.

• VoltAgent awesome-claude-code-subagents — A curated collection of 100+ specialist subagent definitions maintained by the VoltAgent GitHub organisation. These are prompt templates that Claude Code can use to spawn focused helper agents for specific tasks (SQL, frontend, infra, QA). VoltAgent itself is a separate TypeScript framework — we use their subagent catalog, not the framework.

• UI/UX Pro Max — A Claude Code Skill (not a plugin, not an MCP server) that provides a searchable database of UI styles, colour palettes, font pairings, chart recommendations, and UX guidelines. It enhances Claude's ability to make good design decisions.

• Claude in Chrome — An official Anthropic browser extension that lets Claude navigate, read, click, and interact with web pages. In our workflow, it is used exclusively for verifying that the UI we build works correctly in the browser. It is a developer tool, not a general browsing assistant.

• Claude Code GitHub Action — An official GitHub Action (anthropics/claude-code-action) that integrates Claude into your CI/CD pipeline. It can review PRs, respond to @claude mentions in issues, and automate issue-to-PR workflows.

• Serena - An MCP server by Oraios AI that provides Language Server Protocol (LSP) intelligence to Claude Code. Gives Claude IDE-level capabilities: go-to-definition, find-all-references, type hierarchies, and symbol-aware code navigation across 30+ programming languages. Installed per-project using the claude-code context to avoid duplicating Claude Code's built-in file operations. Think of it as giving Claude the same code understanding that VS Code or JetBrains have natively.

• obra/superpowers - A Claude Code plugin (widely adopted, in the official Anthropic marketplace) by Jesse Vincent that provides a development methodology layer. Contains skills that auto-activate based on task context: brainstorming before code, plan-then-execute workflow, test-driven development, systematic debugging, and verification before completion. Token-light design loads ~100 tokens at session start. Complements project-specific rules by encoding HOW to approach development tasks systematically.

• code-review plugin - An official Anthropic Claude Code plugin that runs 5 parallel Sonnet review agents on PR changes. Analyses CLAUDE.md compliance, bug detection, git history context, previous PR comments, and code comment accuracy. Findings are confidence-scored (0-100) with an 80+ threshold to filter false positives. Invoked via /code-review in the terminal or /code-review --comment to post findings to a GitHub PR.

• pr-review-toolkit plugin - An official Anthropic Claude Code plugin providing 6 specialised review agents: comment-analyzer, pr-test-analyzer, silent-failure-hunter, type-design-analyzer, code-reviewer, and code-simplifier. Unlike the code-review plugin (which runs all agents at once), these activate based on natural language prompts, allowing targeted investigation of specific concerns. Particularly valuable for catching swallowed errors and test coverage gaps.

• GitHub MCP Server - GitHub's official MCP server that gives Claude structured API access to GitHub: read/create issues, manage PRs, review code, search across repos, and read CI/CD logs. Connects via remote HTTP endpoint with a Personal Access Token. Separate from the gh CLI (basic operations) and the Claude Code GitHub Action (CI/CD automation). Installed at user level so the PAT is never committed to project files.

• frontend-design (Anthropic official) - An official Anthropic skill (in their claude-code-plugins repository) focused on producing distinctive, bold frontend interfaces that avoid "AI slop" - the generic, safe, homogeneous aesthetics that most AI-generated UIs produce. Opinionated and creative-first. Before writing code, it forces Claude to commit to a bold aesthetic direction (brutalist, luxury, retro-futuristic, organic, editorial, etc.) and then execute that vision with precision. It explicitly bans generic patterns: Inter/Roboto fonts, purple gradients on white, predictable layouts, default Tailwind colours.

Protocols and Configuration

• MCP (Model Context Protocol) — A standard protocol for connecting AI models to external tools and data sources. MCP servers provide capabilities (like documentation lookup or database access) that Claude can use during a session. Configured via .mcp.json or claude mcp CLI commands.

• .mcp.json — A JSON file at the project root that lists the MCP servers this project uses. This is the ONLY correct place for project-level MCP configuration. Do NOT put MCP config in .claude/settings.json — it will be silently ignored.

• .claude/settings.json — Project-level settings for Claude Code. Controls permissions (allow/deny/ask), environment variables, hooks, and defaults. Committed to git and shared with the team.

• .claude/settings.local.json — Developer-specific settings. Auto-gitignored. Use for local database URLs, personal API keys, or per-developer preferences.

• .claude/rules/\*.md — Modular rule files that Claude loads automatically. Can be path-scoped using YAML frontmatter. Used for enforcing team standards (security, testing, git workflow).

• .claude/commands/\*.md — Custom slash commands. Each file becomes a /project:[filename] command.

• .claude/agents/\*.md — Custom subagent definitions. Each file defines a specialist agent with its own model, tools, and instructions.

Concepts

• ADR (Architecture Decision Record) — A short document recording a significant technical decision: what was decided, why, and what alternatives were considered. Stored in /docs/adr/. The purpose is to answer "why did we build it this way?" months or years later.

• Idea Pack — A structured document capturing a product idea: the problem, solution, target users, user stories, scope, and risks. Created before the PRD.

• PRD (Product Requirements Document) — A detailed specification of what to build, including user stories with acceptance criteria, data models, API contracts, and a phased delivery plan. Created from the Idea Pack.

• Acceptance Criteria (AC) — Specific, testable conditions that must be true for a user story to be considered complete. Written in Given/When/Then format. Example: "Given a logged-in user, when they click 'Delete Account', then their account is deactivated and they are logged out."

• Phase — A chunk of work that can be built and tested independently. Each phase delivers one or more user stories. We build one phase at a time, never the whole app at once.

• Context Window — The amount of text Claude can "see" at once. It includes your conversation history, CLAUDE.md, loaded files, tool outputs, and more. When it fills up, Claude loses the ability to remember earlier parts of the conversation.

• Context Compaction — The process of summarising conversation history to free up context space. Done manually with /compactor automatically when context reaches ~85%.

• Subagent — A separate Claude Code instance spawned to handle a focused task. It has its own context window and reports results back to the parent. Used for isolated, well-defined pieces of work.

Abbreviations