Cheat sheet
Full Lesson Reference
One-page reference of everything you'll use daily. Print this. Pin it near your workspace. Come back whenever you forget a command or the right prompt.
Starting + ending sessions
- claude - start a session
- claude --continue (or -c) - resume the most recent session in this folder
- claude --resume - pick a past session from a list (arrow keys + Enter)
- /startup - load project memory + context
- /wrapup - end session (save + commit + push)
- /mcp - list active MCPs
- /model - switch between Opus and Sonnet mid-session
- / (just the slash) - list all installed skills
- /caveman / /caveman off - toggle compressed responses
Permission prompts
- 1 - Yes, just this once
- 2 - Yes, don't ask again this session
- 3 - Yes, don't ask again in this project
- 4 - No, tell Claude what to do differently
- Escape - abort the prompt entirely
- Ctrl+C - kill the session
Context optimisation
Ask every 5-10 messages on long sessions
How full is the context window right now? Rough percentage.
When you need to see exactly what's eating tokens
Before I type anything else - show me exactly what's loaded in this session and the approximate token cost of each:
- CLAUDE.md files (global + project)
- Skills (each with metadata size)
- MCPs (each with connection status + approximate schema cost)
When context is getting heavy and you want to keep working
Summarise what we've discussed in one paragraph. I want to start fresh from this summary.
Then /wrapup + close terminal + open new session + /startup. New session picks up from the summary without dragging the back-and-forth tokens.
When you've pulled a file that was too big
Read pages 3-5 of this PDF (executive summary + conclusions). Summarise.
Read rows 1-100 of this CSV. Read the first 200 lines of this transcript.
Session + wrapup prompts
Start every session the same way
terminal/startup
Today I want to: finish the weekly report for [client], draft 3 emails for the promo campaign, and review the Google Ads account for waste.
Before you wrap, force a clean summary
Summarise what we accomplished today. Then list everything that's still pending - not finished, postponed, or blocked. Group by priority.
For long, complex sessions where the hand-off matters
Before we /wrapup - I'm mid-flow on [complex task]. Write a VERY detailed summary of what we've done so far, what the next specific steps are, and any decisions I've made that might otherwise get lost. Then /wrapup with that summary.
The "capture decisions + reasoning" wrapup prompt
Before you wrap up - summarise the key decisions we made and the reasoning behind each. List anything I should remember for next session even if it feels obvious. Then run /wrapup with that context.
Memory + recall prompts
Search my session_memories for any session where we worked on [topic]. Show me the summaries.
When did we decide to [X]? Find that session and summarise the reasoning.
Read the last 3 session summaries for this project. What's the full arc? What's still outstanding?
Show me the last 5 sessions on [project].
Add to the project record for [key]: [fact]. [fact]. [fact].
Save this as a rule in global_rules: [rule]. Category: feedback (or pattern / reference).
CLAUDE.md management
Draft my global CLAUDE.md at ~/.claude/CLAUDE.md. Ask me 5-10 questions first.
Audit my global CLAUDE.md. For each line: does it apply to every project? Would removing it cause mistakes? Flag for deletion, movement, or keep. Target: under 25 lines.
Show me which CLAUDE.md files are loaded + line counts.
For every rule in my CLAUDE.md that mentions a specific client or project, move it to that project's CLAUDE.md instead.
Skills
Install [skill-name] from [repo-url]. After install, confirm the trigger works.
I'm attaching [N] SKILL.md files. Install each to ~/.claude/skills/[skill-name]/SKILL.md and confirm all trigger as slash commands.
Audit my skills. Report usage vs 16K budget. Flag skills I haven't used in 60+ days. List descriptions over 400 characters.
Disable the [skill-name] skill so it stops loading in new sessions. I might want it back later.
Update the [skill-name] skill from its source repo. Back up my local version first in case I've customised it.
Edit the [skill-name] SKILL.md to also do [specific thing]. Show me the before and after.
MCPs
Install the [platform] MCP. Walk me through auth.
What MCPs are active in this session?
Disconnect the [platform] MCP for the rest of this session. I only need [other platform] today.
Install the [platform] MCP for this session only. Don't add it to .mcp.json - just connect for now.
Install the Google Ads MCP in read-only mode. I don't want Claude to be able to make mutations yet. List all connected MCPs in this session. For each, show the approximate schema size + whether I've invoked it in the last week. Flag any I should disconnect.
Check if enableAllProjectMcpServers is true in any of my project settings. Change to false and list only the servers I actually use.
Planning + brainstorming
terminal/brainstorming
I want to build [goal]. Help me clarify scope before I jump into execution.
Without the superpowers skill
I want to [describe goal]. Help me brainstorm the scope. Ask me 10 questions one at a time that will force me to clarify what I'm actually trying to do. Don't propose solutions yet
- just help me think clearly about the problem.
Turn the brainstorm into a brief
Based on my answers, write a brief.md that captures the goal, success criteria, constraints, known unknowns, and rough output shape. Save to the project folder.
Draft the plan
terminal/writing-plans
Read the brief.md in this project. Draft a plan.md that breaks the work into 4-6 phases, each with steps, dependencies, rough time estimates, and risks.
Update as you execute
Update plan.md - mark Phase 1 complete. Note what went as planned and what surprised us. Adjust remaining phases if our learnings should change them.
Use @ syntax to pull context on-demand
Let's work on Phase 3 from @plan.md. Make sure anything we build matches the constraints in @brief.md.
QA + verification
Source-trace every number
For every metric in this deliverable, trace it back to source. Which database query, MCP pull, or file did each number come from? Flag anything that traces only to "Claude calculated" or "inferred from context".
Force Claude to self-critique
Read this deliverable critically. Flag any claim, number, statistic, or recommendation that you can't verify from the data you actually pulled this session. Be honest about what's made up vs what's sourced.
Cross-section consistency
Check this deliverable for internal consistency. Do the numbers in section 1 match what's referenced in section 3? Does the headline finding align with the detailed analysis? Flag any contradictions.
Responsive check
Review this HTML for responsive design issues. Check it renders properly at 375px (phone), 768px (tablet), and 1440px (desktop). Flag any layout breakage at each size.
Brand consistency
Compare this deliverable against the brand guide at @brand-guide.md (or the client's project CLAUDE.md). Flag any colour, font, logo, voice, or tone inconsistencies.
Wrong-client catch
Search this file for any client names that aren't [correct client name]. Flag anything that looks like it was copied from another project or template.
Date-range check
Confirm the date range for all data in this report. Is everything from [period I expected]? Show me the dates used in each query.
Apply a deliverable checklist
Apply the [report / audit / landing page / email / proposal / ad] QA checklist to this deliverable. Go through each item, flag anything that doesn't pass, and show me the specific issue.
Pre-commit secret scan
Before this commit - scan every staged file for anything that looks like an API key, token, password, or credential. Show me what was found.
Token-saving setup (one-time)
Install RTK from rtk-ai.app and set it up for this terminal.
Install the Caveman plugin from JuliusBrussee/caveman.
Update my /startup and /wrapup skills so they explicitly bypass RTK compression (use rtk proxy on any curl calls) and turn Caveman off at the start of the skill then restore the previous mode at the end. Memory load + save must be full-fidelity, not compressed.
Check savings
rtk gain # cumulative token savings
Switching between Opus and Sonnet
Biggest single optimisation - pick the right model for the job. Opus is smartest + most expensive. Sonnet is ~5x cheaper + still excellent.
/model # see available models + switch
Use Opus for: planning, strategy, heavy analysis, client-facing writing that has to land, first drafts of skills or voice guides.
Use Sonnet for: file moves, git commits, MCP data pulls, formatting conversions, simple typo fix es, following an already-written plan, installs + audits + status checks.
Start: Opus (brainstorm, plan, strategy)
Execute: Sonnet (follow the plan)
Stuck: Opus (break the blocker)
Finish: Sonnet (commit, push, wrapup)
Default Sonnet, escalate to Opus - not the other way round. Saves 40-60% of daily token usage with no quality hit.
Troubleshooting + state checks
What folder are you running in?
Show me active MCPs + their connection status.
Show me which CLAUDE.md files are loaded + line counts.
Test my Supabase memory connection.
How full is the context right now?
I just got this error. Help me fix it.
Confirm my .env has [KEY_NAME] set to a non-empty value. Don't print the value - just confirm it's set.
Work commands
Build me a [weekly report / landing page / email / audit] for [client].
Pull the last [X days] of [platform] performance.
Build a weekly report. Save to drafts/weekly-report-[date].html.
Commit and push everything we changed this session.
Push this file to my Pages repo + give me the live URL.
This is done. Commit everything and push to GitHub so it goes live.
Correction prompts when output is off
That's not what I meant. Try it with [specific change]. The tone is off - make it more direct. Cut section 3 and rebuild section 2 with this data instead.
File locations
- ~/.claude/CLAUDE.md - global CLAUDE.md
- ~/.claude/skills/ - installed skills
- [project]/CLAUDE.md - project-level CLAUDE.md
- [project]/.env - API keys (gitignored)
- [project]/.mcp.json - MCP config for this project
- [project]/brief.md - project brief
- [project]/plan.md - execution plan
- [project]/notes.md - decision log
Safety rules (always-on)
- NEVER paste API keys into prompts - put them in .env via your text editor
- NEVER approve ad mutations without a written plan reviewed
- NEVER approve email sends without reading full content + segment
- NEVER commit files with credentials
- NEVER force push to main
- NEVER drop database tables (rename with archive prefix)
The session loop
- Right-click project folder → Open in Terminal
- Type claude
- Type /startup
- Tell Claude what you want in plain English
- Answer permission prompts with number keys
- Ask "how full is context?" every 5-10 messages on long sessions
- At session end: /wrapup
- Close the terminal tab
The monthly audit loop
30 minutes, once a month
- Baseline audit - what's loaded, what's the token cost
- CLAUDE.md audit - under 25 lines global, 30-60 project
- Skill audit - under 16K budget, remove unused
- MCP audit - disconnect what you're not using
- Memory audit - archive finished projects
- Folder audit - anything to reorganise?
Installation quick links
- Claude Code native install (Mac): curl -fsSL https://claude.ai/install.sh | bash
- GitHub CLI: brew install gh (Mac) / cli.github.com (Windows)
- Node.js: nodejs.org (LTS version)
- ccstatusline (live status bar): npx ccstatusline@latest
- RTK: curl -fsSL https://www.rtk-ai.app/install.sh | sh
- Caveman: claude plugin marketplace add JuliusBrussee/caveman && claude plugin install caveman@caveman
- Superpowers pack: github.com/obra/superpowers
When it's broken
- Is CLAUDE.md too long or contradictory?
- Is context over 50%?
- Did you run /startup?
- Are you in the right folder?
- Fresh session + /wrapup + restart
Action items
☐ Print this page OR bookmark it
☐ Pin near your workspace
☐ Reference when you forget commands or the right prompt
Module complete. Course complete. You've finished the Claude Code Training classroom. You now have the full toolkit - from terminal setup through advanced workflow patterns, safety, and quality discipline. The rest is practice.
Exercises
- Review the concepts covered in this lesson: Cheat sheet.
- Write down your key takeaway from this lesson.
- Practice running any commands or prompts mentioned above inside your terminal.