# AI Teachers — Build With Codex and Claude Code Updated 2026-10-25 · Free written training workbook Use invented or permitted data. Your AI tool may require a separate subscription. Download this workbook and save answers locally; this course site saves only completion checkmarks. ## The outcome Learn a disciplined AI-assisted software workflow from requirements and a small repository change to tests, review and a cautious release. ## What you will learn - Define an achievable feature - Brief a coding agent safely - Inspect code diffs and tests - Check accessibility and quality - Prepare release and rollback ## Lesson 1: Choose a tiny project (30 min) Small verifiable changes are the fastest way to learn engineering with agents. ### Follow these steps 1. Start with a sample project or a repository you own and can edit. 2. Choose one feature: a reading-list checklist with local persistence. 3. Write acceptance criteria for normal, empty and incorrect input. 4. Set the scope: new branch only, no production changes, no secrets. ### Copyable prompt Act as a senior engineer helping a beginner. I want a reading-list checklist that can add an item, mark it done, and persist locally. Write a one-page specification with scope, acceptance criteria, edge cases, accessibility requirements, test cases and a list of files we should inspect before editing. Do not write code yet. ### Deliverable A concise, testable feature specification. ### Quality gate A second person could decide whether the feature passes. ### Watch out Avoid using a production repository or deploying on lesson one. My result: What I verified: What I will change: --- ## Lesson 2: Inspect the codebase and plan (35 min) Before editing, understand existing conventions and where the change belongs. ### Follow these steps 1. In Codex or Claude Code, open the safe practice repository and check the active branch. 2. Ask for a map of relevant components, data flow and existing tests. 3. Request an implementation plan with risks and a minimal diff. 4. Review proposed file edits before granting any write permission. ### Copyable prompt Inspect this practice repository in read-only mode. Find the entry point, related UI component, state handling, test setup and build command. Propose a minimal implementation plan and explain which files would change, why, and how to test. Do not modify files until I explicitly approve the plan. ### Deliverable Repository map and file-level plan. ### Quality gate The proposed approach matches the project's actual stack. ### Watch out AI-generated file paths and commands must be verified inside the real repository. My result: What I verified: What I will change: --- ## Lesson 3: Build on a branch and run tests (45 min) An agent can generate code; shipping it is still a human engineering decision. ### Follow these steps 1. Create a feature branch and confirm a clean worktree. 2. Implement only the approved acceptance criteria. 3. Run the existing test suite and add tests for empty inputs and persistence. 4. Inspect the diff for unrelated edits, dependency surprises and exposed secrets. ### Copyable prompt Implement the approved reading-list feature on the current feature branch only. Keep the diff minimal. Add tests for add, complete, empty-input and local-persistence behavior. Run the project's test and lint commands. Report exactly what ran, with pass/fail results and any limitations. Do not commit to main, install unknown dependencies, or deploy. ### Deliverable An inspectable patch and actual test report. ### Quality gate Tests actually executed; their output and limitations are recorded. ### Watch out A claimed successful test without tool output is not evidence. My result: What I verified: What I will change: --- ## Lesson 4: Review UX, accessibility and security (40 min) Functional code may still be confusing, inaccessible, or unsafe. ### Follow these steps 1. Inspect the feature at phone and desktop widths. 2. Use keyboard-only navigation and verify focus, labels and statuses. 3. Test error, empty and refresh states using real browser behavior. 4. Ask a second review pass to spot data exposure or unnecessarily complex logic. ### Copyable prompt Review the patch as a skeptical accessibility and security reviewer. Check focus visibility, labels, live updates, contrast, empty states, dependency changes, dangerous operations, and secret handling. List issues by severity and specific evidence. Do not claim browser tests unless they ran. ### Deliverable A prioritized defect list, corrected patch and manual QA log. ### Quality gate No critical accessibility or security issue is knowingly left open. ### Watch out Automated test passing does not substitute for real interaction testing. My result: What I verified: What I will change: --- ## Lesson 5: Release only after approval (40 min) Finish with a deploy plan and a path back to safety. ### Follow these steps 1. Prepare a pull request with summary, screenshots, tests and unresolved risks. 2. Verify deployment configuration, backups and rollback instructions. 3. Request explicit approval before merging or publishing. 4. After release, check the live feature and document issues for follow-up. ### Copyable prompt Prepare a release checklist and PR description for this change. Include acceptance criteria, exact tests run, accessibility checks, changed files, remaining risks, release owner, rollback command or procedure, and post-release smoke tests. Do not merge or deploy. ### Deliverable A release-ready review package with rollback steps. ### Quality gate The release owner can confidently approve or reject it. ### Watch out Never disclose API keys, disable guardrails or pretend a deployment happened. My result: What I verified: What I will change: --- ## Final review - [ ] Every material fact has a permitted source. - [ ] Missing data is visible. - [ ] External actions still require authorized approvals. - [ ] I have tested at least one failure case. - [ ] I know who owns the result and how to stop the workflow. Official reference links: - ChatGPT Work and Codex: https://help.openai.com/en/articles/20001275-chatgpt-work-and-codex - Claude Code documentation: https://docs.claude.com/en/docs/claude-code/overview AI Teachers · https://aiteachers.cc/academy/codex-claude-code/