Sefism Coding with AI Tools Cheat Sheet

What this cheat sheet is

The Coding with AI tools cheat sheet is the companion to the Coding with AI tools roadmap, for students and developers who use an AI coding tool and want to stay the engineer in charge. It covers what the tools are good and bad at, project instructions, prompting, reviewing, testing, debugging, agent safety, security and keeping your own skills. Use it to look up a command or setting, and to check your habits before you trust a change.

Who it is for

Anyone who already has a tool installed and uses it daily: students working on assignments and projects, interns figuring out what their team allows, and developers whose company has rolled out an assistant. It is short by design, so it fits next to your editor.

Editions

There are three editions: Claude Code, Cursor and GitHub Copilot. Each has the same ten chapters and 77 concepts. The first chapter includes an entry on your tool, and the setup, permissions and safety entries use your tool's own files, commands and controls.

Chapter by chapter

1. What AI coding tools are good and bad at. Large language models, hallucination, the context window, completion, chat and agent, where AI helps most, where it struggles, automation bias, and your tool.
2. Setup and project context (8 concepts). Installing and starting your tool, the project instructions file, AGENTS.md, generating a starter file, path-scoped rules, keeping files out of context, what belongs in instructions, and pointing at exact context.
3. Prompting for code (7 concepts). Small, checkable tasks, a task spec, pointing at an existing pattern, plan before code, prompt templates, saved prompts and skills, and fresh context per task.
4. Reading and reviewing AI code (7 concepts). Read every diff, a review checklist, explain it back, checking invented APIs, unrequested changes, commit before each AI task, and AI review as a second pass.
5. Tests first, so AI output is checked (8 concepts). Tests as the spec, red, green, refactor, writing the first test yourself, tests that cannot fail, an edge case list, giving the agent a check to run, not letting it edit tests to pass, and mutation testing.
6. Debugging with AI (7 concepts). Reproduce before you ask, minimal reproduction, hypotheses not fixes, paste the evidence, symptom patches, root cause and regression test, and the real debugger.
7. Agents: multi-file changes, permissions and safety (9 concepts). The agent loop, permission modes, allow and deny rules, checkpoints and undo, Git as the safety net, sandboxing, prompt injection, MCP servers, and background and cloud agents.
8. Security and licences (9 concepts). Never paste this, secrets out of code and context, hallucinated packages, checking a new dependency, insecure patterns AI repeats, static analysis, licences and copied code, data use and privacy settings, and employer and university AI policy.
9. Keeping your skills (7 concepts). Deliberate practice, AI-off sessions, attempt first then compare, tutor mode prompts, read the docs first, fundamentals AI does not replace, and explaining without notes.
10. Shipping a real project with an agent (7 concepts). Scoping for an agent, a written spec, vertical slices, CI as the gate, one pull request per slice, disclosing AI use, and the project write-up.

Rules worth pinning

The sheet's tips are short enough to stick on your monitor:

• If you cannot explain a line, do not merge it.
• Commit before every AI task, so any change is one command away from gone.
• Give the tool a check to run and ask for the output, not a promise.
• Small tasks, small diffs: if the diff is too long to read, the task was too big.
• Write the first test yourself; that is where you decide what the code should do.
• Verify every function, flag and package name against the official docs.
• After two failed corrections, clear the context and write a better prompt.
• Practise with AI off every week, then compare your answer with the tool's.

How to use it with the roadmap

• Before each working session, glance at chapters 3 and 4: one task, one spec, one reviewed diff.
• When you set up a new repository, go through chapters 2, 7 and 8 together: instructions file, permissions and secrets before the first prompt.
• Once a month, reread chapter 9 and check you still have AI-off practice in your week.

How it works on Sefism

Everything on Sefism is free to learn from. The public page at Coding with AI tools cheat sheet shows every chapter title, with the first two chapters open in full. Later chapters show their title, a short introduction and how many concepts they hold.

Keeping the sheet uses credits, and it is yours for good after that. Adding the Coding with AI tools roadmap to your dashboard also opens this sheet, so if you plan to follow the roadmap you do not need to keep the sheet separately. New members get starting credits when they join, and the first roadmap is free. Create a free account to get started, and if a concept still does not make sense after reading it, ask Tauseef a question.

Related on Sefism

• How to check if AI-generated code is correct
• How to get better at debugging without asking AI first
• Building Effective Agents, for how the agent loop works
• The MCP cheat sheet, if chapter 7's MCP servers interest you

Frequently asked questions

Is this a list of keyboard shortcuts? Partly: setup and permissions entries name your tool's files and controls. Mostly it is the habits that make AI output safe to merge.

Which edition should I pick? The tool you use. Concepts on review, testing, debugging and skills are the same in all three.

Does it cover security? A whole chapter: what never goes into a prompt, hallucinated packages, insecure patterns AI repeats, static analysis, licences and AI policies.