Shell WalkerIntermediate10 daysFree
Shipping Safe Code with AI Coding Assistants
Use AI coding tools every day and still ship code you can trust.
Taught by Mr Elite 2 of 10 days live A new day every day
I built this course for developers who already use AI coding assistants and want to stop guessing about the risks. Over ten days I will show you how these tools fail, how attackers take advantage of them, and what habits and settings keep your code and your secrets safe. We cover reviewing AI-written code, secrets in prompts, agent permissions, poisoned issues and package hallucination, which is when a model invents a package name that an attacker can then register.
Each day is one lesson of about 45 to 75 minutes, including a mission you finish before the next day. Most days use a SecurityElites browser lab, and every hands-on step runs against your own machine, a local test setup or the named practice lab. Modules end with a day where you apply what you learned, and from day 6 onwards the module-ending days also review earlier material. Day 10 is a capstone that pulls it all together.
By the end you can
- Review an AI-generated pull request and find insecure code, hidden flaws and tests that prove nothing.
- Spot and redact secrets before they enter a prompt or context window.
- Set least-privilege permissions for a coding agent and explain its blast radius.
- Recognise a poisoned issue or document that tries to hijack an agent, and apply controls against it.
- Verify a package an AI suggests before installing it, and defend against slopsquatting.
- Review an MCP server and its tools before connecting them to your agent.
- Write a one-page AI coding policy for a team, backed by checks you have tested.
Who this is for
- Developers who use AI coding assistants or coding agents in their daily work.
- Learners who are comfortable in a terminal and understand how the web and networks work.
- People who have tried a few security labs and want to apply that knowledge to AI-assisted development.
- Team members who need to set sensible rules for AI tools in their projects.
Before you start
- Comfort using a terminal, git and a code editor.
- Basic understanding of HTTP, web apps and how packages are installed (npm or pip).
- A few labs completed on any platform is helpful but not required.
- No paid AI tool is needed. Any assistant you already use will do for the missions.
- A modern web browser for the SecurityElites labs
- A terminal (Linux, macOS or Windows with WSL)
- Git
- Node.js with npm, or Python with pip
- A code editor such as VS Code
- Any AI coding assistant you already use (free tier is fine)
Syllabus
One day at a time. Finished days get a tick. Days that are not out yet show when they unlock.
Module 1 Foundations: Trust and Review
Learn why AI code needs the same scrutiny as any untrusted contribution and practise reviewing it.
- 1Orientation: Why AI Code Needs ReviewBy the end of today you can explain the main ways AI coding tools put code at risk and complete your first lab.Lab: Tests That Test Nothing
- 2Reading AI Pull Requests Like an AttackerBy the end of today you can review an AI-written pull request using a repeatable security checklist.Practice | Lab: Review the AI's Pull Request
- 3Module Practice: Catching Insecure Generated CodeBy the end of today you can find and fix a classic injection flaw in code an AI assistant wrote for you.Practice | Lab: SQL Injection | Unlocks 17 Nov
Module 2 Protecting Secrets and Supply Chain
Keep secrets out of prompts and stop invented or malicious packages from reaching your project.
- 4Secrets in the Context WindowBy the end of today you can find secrets in material you are about to paste into a prompt and redact them safely.Lab: Secrets in the Context Window | Unlocks 18 Nov
- 5Package Hallucination and SlopsquattingBy the end of today you can check whether a package an AI suggests is real, trusted and safe to install.Practice | Lab: AI Package Hallucination | Unlocks 19 Nov
Module 3 Controlling Agents
Limit what coding agents can do and defend them against hostile input and tools.
- 6Agent Permissions and Blast RadiusBy the end of today you can configure a coding agent with least privilege and explain what a compromised agent could reach.Practice | Lab: Coding Agent Blast Radius | Unlocks 20 Nov
- 7Poisoned Issues and Hostile InputBy the end of today you can show how text in an issue or document can hijack a coding agent and name two controls that reduce the risk.Practice | Lab: Poisoned GitHub Issue | Lab: AI Indirect Prompt Injection | Unlocks 21 Nov
- 8Module Review: Tools, MCP Servers and ApprovalsBy the end of today you can review an MCP server's tools before connecting it and recall earlier checks on code, secrets, packages and permissions.Review | Lab: Rogue MCP Server | Unlocks 22 Nov
Module 4 Putting It Together
Combine every control into a working policy and audit a whole project.
- 9Writing a Team Policy for AI Coding ToolsBy the end of today you can write a one-page policy that turns the course controls into rules a team can follow and check.Practice | Lab: AI Data Exfiltration | Unlocks 23 Nov
- 10Capstone: Audit an AI-Assisted ProjectBy the end of today you can audit a project built with AI help and deliver findings and a hardened configuration.Capstone | Lab: Review the AI's Pull Request | Lab: Poisoned GitHub Issue | Unlocks 24 Nov
The final project
You will audit a small project that an AI assistant helped write, covering code flaws, leaked secrets, package risk and agent permissions. You then deliver a written report and a hardened configuration that you could hand to a team.
Questions people ask
Do I need a paid AI coding tool for this course?
No. The labs run in your browser, and the missions work with any assistant you already use, including free tiers. You can also do the review tasks by hand.
Is this course about attacking AI tools?
It shows you how attacks work so you can spot and stop them. Every hands-on step runs on your own machine, a local setup or a named practice lab, never a real third-party system.
Will the advice still apply if I change tools?
Yes. The checks focus on habits and permissions that apply to any assistant or agent, such as reviewing diffs, redacting secrets and limiting access.
How much time should I plan each day?
Plan 45 to 75 minutes. That covers the lesson, the lab and the mission, and you should finish the mission before you start the next day.