Level up your Gemini CLI - D&D Style!

You've heard about it, people talk about it, you might have tried it a few times too.
But how many XP did you do on CLI? How many hobgoblins do you need to kill, MCP / Extensions do you need to write before getting to L36?

Riccardo is going to walk you through L1 to L36 in pure 1982’s D&D style, from zero to hero and maybe even to Immortal! Don’t sweat on the slides: there will be DEMOS to show you how to progress and refactor your logic from simple to increasingly complex scenarios.

Note: this abstract was (astonishingly) not written by AI, but by a Googler’s brain trained on multiple AI abstracts.

Transcript

1. Google Cloud Proprietary & Confidential 1 Level up your Gemini CLI! Riccardo Carlesso Developer Advocate Zurich, 13 Oct 2025 go/ricc-gemini-cli-slides-dnd
2. 🐦 @SREccardo www.linkedin.com/in/riccardocarlesso/ Former sysadmin, Ruby [on Rails] developer, network admin. 13-years Googler, passionate about Reliability, SRE, DevOps, Maths, yellow, 🎹 piano, triathlons 󰖱🏃󰚬, caipirinhas and Amarone. Bash aficionado. As a Developer🥑Advocate for Google Cloud, I preach SRE, DevOps, the Google Way of doing things to various audiences. About me
3. Google Cloud Proprietary & Confidential 3 Gemini CLI: Basic • Installation • First usage
4. Proprietary + Confidential Open Source! https://github.com/google-gemini/gemini-cli
5. Proprietary + Confidential Proprietary + Confidential Gemini CLI is a locally-run agentic application that orchestrates specialized agents
6. Proprietary + Confidential Google Cloud Proprietary & Confidential 6 Authentication: /auth Five ways to authenticate them all..
7. Proprietary and Confidential Interactive and File/Shell Integration / commands @ context ! shell
8. Proprietary and Confidential Powerful Built-In Tools File system tools (read_file, write_file, list_directory, search_file_content), Shell tool (run_shell_command) Web tools (web_fetch, google_web_search)
9. Proprietary and Confidential Hierarchical Context & Memory Project-specific and global instructions to the AI using GEMINI.md files
10. Google Cloud Proprietary & Confidential /help is your friend: 10 Tip: /help
11. Proprietary + Confidential Google Cloud Proprietary & Confidential 11 Instructions: GEMINI.md Samples for inspiration: https://github.com/google-gemini/gemini-cli/blob/main/GEMINI.md https://github.com/palladius/vibecoding/blob/main/GEMINI.md • GEMINI.md loaded from local folder when you start gemini-cli. • Used for system instructions at the beginning of any chat. Use it! • More tips later.
12. Proprietary + Confidential Google Cloud Proprietary & Confidential Location: code ~/git/gemini-cli-demos/demos/a-repo-from-a-friend/ Demo: DEMO-L1.md Demonstrate : • Code understanding • /help • /memory add Remember to address me as Riccardo McDemo, the best demoer of the west! Add Greek emojis (flag, temples, ..) to our conversations • /memory refresh • Explain the Security feature, which you can bypass with --yolo 12 Demo L1: code understanding
13. Proprietary + Confidential Google Cloud Proprietary & Confidential 13 Demo L2: “git magic” Location: code ~/git/rubycon/ code doc/demos/202510-mythos.md Git investigation Look at the last 30 commits. Show me a table of all committers, 1 committer per row with N commits and try to look at some git diffs to tell me what these people are usually contributing to - is there a pattern?
14. Proprietary + Confidential Google Cloud Proprietary & Confidential 14 Demo L2: “git magic” Location: code ~/git/rubycon/ code doc/demos/202510-mythos.md Git investigation Look at the last 30 commits. Show me a table of all committers, 1 committer per row with N commits and try to look at some git diffs to tell me what these people are usually contributing to - is there a pattern?
15. Google Cloud Proprietary & Confidential 15 Gemini CLI: Expert • Add MCP to the mix. Try context7, ChromeTools, … • Install some Custom Commands via Extensions. • Write many GEMINI.md , in different folders. • Maintain your own pet ~/.gemini/GEMINI.md • Have gemini-cli write a PLAN.md , review it, and have it execute it. ❏ Bonus if it has a checklist. • Start abstracting files in smaller chunks of context in other markdowns: TESTING.md, CONTRIBUTING.md, ...
16. Google Cloud Proprietary & Confidential 16 • GEMINI.md used for LLM instructions. Use it! • Keep it fat! ◦ All your suggestions within the chat will be lost when the program dies, unless you use memory. ◦ You can include others with @GEM…. And autocomplete. • DRY it, and use it everywhere (I use symlinks - maybe a curl in the future if private?) Fat model? => Fat GEMINI.md
17. Google Cloud Proprietary & Confidential 17 Hierarchical GEMINI.mds! Also: ~/.gemini/GEMINI.md ⇐ System Memory (free) ⇒ works with /memory <remember blah blah blah> Your local repo GEMINI.md ..
18. Google Cloud Proprietary & Confidential 18 Enter MCP! More info here: https://github.com/google-gemini/gemini-cli/blob/main/docs/cli/tutorials.md "get all open issues assigned to me in the 'foo/bar' repo and prioritize them" “Navigate to URL and take a screenshot of footer”
19. My fav MCPs Docs: Context7 Browser: Playwright, Chrome DevTools GCP: CloudRun Media: Vertex AI GenMedia (wow!). HouseGemacht: My own Rails app exports ActiveRecord with ruby_llm +fastmcp built-in MCP support. Database: GenAI ToolBox
20. Google Cloud Proprietary & Confidential /memory saves local important state 20 Feature: /memory
21. Proprietary + Confidential Google Cloud Proprietary & Confidential /mcp Show asking questions to an MCP Folder: code ~/git/gemini-cli-demos/demos/nanobanana Open DEMO-L14.md 21 Demo L14: MCP to Nano🍌
22. Google Cloud Proprietary & Confidential 22 Gemini CLI: Companion • Abstract chunks of context into custom commands and install through extensions • Configure GitHub CI/CD with /setup-github
23. Google Cloud Proprietary & Confidential 23 Custom commands • Is your GEMINI.md too bloated? Does it have info about your app, how to manage the DB, CICD, Google Cloud, and git commits? Does it confuse info from one into the other? ◦ => time to abstract these into sub-commands. https://github.com/palladius/gemini- cli-demos/tree/main/.gemini/comma nds/common
24. Google Cloud Proprietary & Confidential 24 Supports Extensions! • Gemini CLI supports extensions that can be used to configure and extend its functionality. • Use gemini-extension.json • Same as apt-get install functionality More: https://github.com/google-gemini/gemini-cli/blob/main/docs/extensions/index.md
25. Google Cloud Proprietary & Confidential https://geminicli.com/extensions/browse/ 25 geminicli.com > extensions > browse
26. Google Cloud Proprietary & Confidential 26 github actions
27. Proprietary + Confidential Google Cloud Proprietary & Confidential All on Web Show the App: https://ideas-app-dtkq332qua-ew.a.run.app/ Create an app for managing monthly meetings with announcements, email sending, collecting feedback, … https://github.com/palladius/booth-ideas-sg/ * Issues * Actions / in progress Apps: https://palladius.github.io/booth-ideas-sg/ 27 Demo L25: GitHub Actions
28. Proprietary + Confidential Google Cloud Proprietary & Confidential Location: code ~/git/gemini-cli-demos/demos/a-repo-from-a-friend/ Demo: DEMO26.md Here we create our own Custom Command 28 Demo: L26 Create custom command
29. Google Cloud Proprietary & Confidential 29 Gemini CLI: Master • Use extensions to build different personas. Instantiate different personas in parallel. • Create your own MCP • Add command execution !{} to your custom commands. • Add generic CUJs for Gemini to execute (demo) • Create agentic applications which combine all the above. • Add complex, custom logic, to background activities (GHA, cron jobs, ..)
30. Google Cloud Proprietary & Confidential 30 Does it get confused among your many instructions? Too much context?
31. Google Cloud Proprietary & Confidential 31 Context Eng vs Prompt Eng
32. Google Cloud Proprietary & Confidential 32 Context Rightsizing https://www.anthropic.com/engineering/effective-context-engineering -for-ai-agents
33. Google Cloud Proprietary & Confidential 33 Does it get confused among your many instructions? • Abstract pieces of contents in “gems”. ◦ Put them all in the same folder and tell GC how to find them (1 line in G.md) ◦ Everything about github? => docs/gemini/github.md • Feeling adventurous? Build your custom command using that. ◦ As of 0.4.0, you can also refer your Md directly in the command! ◦ Eg, .gemini/commands/git/my_action.toml can reference @doc/my_git.md! • Do not rely on /memory: it gets messy really soon. Use compaction. ◦ put it on git, maybe with chezmoi. • Try Prashanth’s advanced-gemini-cli-part-2-decoding-the-context Too much context?
34. Proprietary + Confidential Google Cloud Proprietary & Confidential Location: code ~/git/gemini-cli-demos/demos/mcp-video-creation/ Demo: code doc/demos/20251013-mythos-storyboard.md Here we see a CC in action. 34 Demo L36 pt1
35. Google Cloud Proprietary & Confidential 35 Gemini CLI: Immortal • Orchestrate multiple concurrent agents in your machine, using disk as status exchange. • Come up with new, unexisting, unthought of functionality. File a PR. • Reach AGI without wiping out humanity. • UTCP?
36. Proprietary + Confidential Google Cloud Proprietary & Confidential 36 Celestial Idea: Orchestrate N Gemini CLIs
37. Proprietary + Confidential Google Cloud Proprietary & Confidential 37 Empyreal Idea: teach it to think Some ideas: • Look for sequentialthinking MCP • 2386 emergent puzzle solving, uses PROJECT_PLAN.md , ASSUMPTIONS.md , DECISION_LOG.md , ..
38. Proprietary + Confidential Google Cloud Proprietary & Confidential 38 Hierarch: Gemini CLI + Computer Use?
39. Level up your Gemini CLI - D&D Style You've heard about it, people talk about it, you might have tried it a few times too. But how many XP did you do on CLI? How many hobgoblins do you need to kill, MCP / Extensions do you need to write before getting to L36? Riccardo is going to walk you through L1 to L36 in pure 1982’s D&D style, from zero to hero and maybe even to Immortal! Don’t sweat on the slides: there will be DEMOS to show you how to progress and refactor your logic from simple to increasingly complex scenarios. Note: this abstract was (astonishingly) not written by AI, but by a Googler’s brain trained on multiple AI abstracts. Abstract