Glossary
Full Lesson Reference
Quick reference of every important term in this course. Bookmark this lesson - come back when something's unfamiliar.
Core concepts
Claude Code
An AI agent that runs on your machine. Reads files, writes new ones, runs commands. Not a website - software in your terminal.
Session
A single run of Claude Code, from when you type claude until you type /wrapup + close the terminal.
Context window
Claude's working memory for one session. Everything accumulates in it - prompts, responses, files read. Has a limit.
Token
The unit Claude uses to process text. ~0.75 words per token. Every message resends the full conversation, so tokens compound.
CLAUDE.md
A markdown file with instructions for Claude. Global (~/.claude/CLAUDE.md) applies everywhere. Project CLAUDE.md only loads when Claude runs in that folder.
The 4 foundations
Claude Code
The AI agent itself.
Database / memory layer
Supabase project that persists data across sessions. Stores projects, session summaries, global rules.
GitHub
Cloud storage + version history + publishing. Backs up your work, deploys files to live URLs.
Connectors
How Claude talks to platforms - MCPs + CLIs.
Connectors
MCP (Model Context Protocol)
A layer that lets Claude Code talk to external platforms (Google Ads, Meta, Klaviyo, etc.). Structured tool interface.
CLI (Command-Line Interface)
A tool that runs in your terminal and does one thing. gh (GitHub), gws (Google Workspace), supabase. Lean alternative to MCPs.
Claude Connectors
Official Anthropic-provided MCPs. One-click auth.
Pipeboard
A 3rd-party provider for ad platform MCPs (Google Ads, Meta, TikTok).
Supermetrics
Cross-platform data MCP covering 150+ marketing sources through one connection.
.mcp.json
Project-level file listing which MCPs to activate for that project. Scopes MCPs per project.
Skills
Skill
A reusable workflow packaged into a slash command. Lives at ~/.claude/skills/skill-name/SKILL.md.
SKILL.md
The file that defines a skill - frontmatter (trigger + description) + instructions.
/wrapup + /startup
Custom skills (installed from Module 04) that save + load session memory. /wrapup at session end, /startup at session start.
Slash command
How you invoke a skill. Type / followed by the skill name (e.g. /wrapup).
16K skill budget
All skill descriptions combined can't exceed 16,000 characters. Hard limit.
Sessions + memory
Memory layer
Your Supabase database that persists across sessions. Stores projects, session summaries, global rules.
projects table
One row per client or initiative you work on.
session_memories table
One row per completed session.
global_rules table
Cross-project feedback, patterns, references.
The 50% rule
When context window hits 50% full, /wrapup + start fresh. Quality degrades past 50%.
terminal/compact
Built-in feature that compresses context mid-session. DON'T USE - /wrapup + fresh session is better.
Permissions
The 1/2/3 prompt
Numbered options Claude shows before taking an action. 1 = yes, 2 = yes + don't ask this session, 3 = yes + don't ask this project, 4 = no.
Auto mode
Runtime setting that skips routine permission prompts while keeping rails on destructive actions.
Dangerous mode
--dangerously-skip-permissions flag. Removes ALL permission prompts. Never use for real work.
GitHub + publishing
Repo
Folder GitHub tracks. Every file + change versioned.
Commit
Snapshot of your work at a specific moment.
Push
Send commits from your machine to GitHub.
Pages repo
GitHub repo where files in the public/ folder auto-publish to a live URL.
Workspace repo
Private repo for your day-to-day work in progress.
StatiCrypt / encryption
AES-256 encryption for files in your Pages repo. Password-protects client-facing content.
Tokens + performance
Baseline tokens
What loads in every session before you type - CLAUDE.md + skills + MCPs + memory. Target 25-35K.
Deferred loading
MCPs now only fully load schemas when tools are actually called. Reduces idle overhead.
RTK (Rust Token Killer)
Tool that strips verbose terminal output before it counts against context. 20-40% savings on output tokens.
Caveman
Skill that compresses Claude's responses. Toggle with /caveman. 20-50% savings.
ccstatusline
Terminal add-on that puts a live status bar at the bottom of every Claude Code session - context %, current model, git branch, session cost, session time. Install with
terminalnpx ccstatusline@latest. Biggest single win for context awareness (see the 50%
rule).
Context engineering
@file syntax
Reference another file in a prompt or CLAUDE.md. Claude reads the file content as if inlined. brief.md / plan.md / spec.md / notes.md
Canonical context files for complex projects. Brief = goal + constraints. Plan = phases + steps. Spec = detailed requirements. Notes = decision log.
Safety
Prompt injection
Hidden instructions inside content Claude reads. Can hijack a session. Stick to trusted sources.
.env file
Plain text file in project folder holding API keys + credentials. ALWAYS in .gitignore. Never pasted into prompts.
Guard hooks
Automated safety checks that catch dangerous actions before they execute.
Anti-fabrication rule
A rule (in your CLAUDE.md) or custom skill that enforces: every metric traces to a verifiable source. Prevents hallucinated numbers. Not a default feature - you build it.
terminalclaude --continue
Resumes your most recent session in the current folder. Every message + tool call + approval loads back. Use when a session crashed or closed before you hit /wrapup. Shortcut: claude -c.
terminalclaude --resume
Shows a picker of your recent sessions for this folder. Arrow keys + Enter to load a specific one. Use when you want an earlier session back, not just the most recent.
Quick command reference
- claude - start a session
- claude --continue (or -c) - resume the most recent session
- claude --resume - pick a past session from a list
- /startup - load project context from memory
- /wrapup - end session properly (save, commit, push)
- /model - switch Claude models
- /mcp - see active MCP connections
- / - list all available skills
- Escape - interrupt current task
- Ctrl+C - kill the session entirely
- 1/2/3/4 - answer permission prompts
- npx ccstatusline@latest - install or reconfigur e the live status bar (context % / model / cost)
Action items
☐ Bookmark this lesson for reference
☐ Come back whenever you see a term you don't remember
Next lesson: Cheat sheet of daily commands.
Exercises
- Review the concepts covered in this lesson: Glossary.
- Write down your key takeaway from this lesson.
- Practice running any commands or prompts mentioned above inside your terminal.