What this video shows
Tom Chant builds the workflow in stages. He starts with prompts that specify the task, relevant context and expected output, then adds nested AGENTS.md files and reusable skills so an agent can apply project rules consistently.
The second half covers characterization tests, test-driven development, Git hygiene, secret handling, context management, permissions, config.toml and an MCP connection. The course repeatedly asks the learner to inspect diffs and verify behavior instead of accepting generated code on appearance.
Study the exercises in the interactive Scrimba course while following the video. Compare its repository-instruction examples with the AGENTS.md specification. Use OpenAI's Codex configuration reference to check current permission and configuration names.
What you will learn
- State the task, relevant files and acceptance criteria in the prompt.
- Keep persistent repository rules in versioned instruction files and focused skills.
- Inspect diffs and run tests before treating an agent task as complete.
- Use Git checkpoints before risky edits and keep secrets out of prompts and tracked files.
- Grant only the commands, paths and network access required for the task.
How to apply this safely
- Choose one small repository task with an observable acceptance criterion.
- Add or review the nearest AGENTS.md file and remove instructions that do not apply to that subtree.
- Ask the agent for a plan, then confirm the files, tests and commands it may use.
- Write a failing characterization or regression test before the implementation when the seam permits it.
- Review the diff, run the focused and full checks, and commit only the intended files.
Important limitations
- The demonstrations use Scrimba's teaching agent, so interface details differ from Codex, Claude Code and other tools.
- The course introduces MCP setup but does not provide a full threat model for third-party servers.
- A single guided project cannot represent the constraints of every language, repository or deployment pipeline.
- Tool names and configuration options can change after publication, so learners should check current vendor documentation.
Sources to check
- Codex & AI Coding Agents - Learn Agentic Coding freeCodeCamp course by Tom Chant, published November 10, 2026.
- Learn Agentic Coding Interactive Scrimba course and exercises.
- AGENTS.md Open repository-instructions format used across coding agents.
- Codex configuration reference Current OpenAI reference for Codex configuration.
Continue learning on Learnetto
Review AI-generated code
Use an evidence-first review loop before merge.
Sandbox AI coding agents
Limit credentials, files, commands and network access.
Build an AI agent test set
Turn real tasks into repeatable acceptance checks.