# Codex masterclass: organise a project, run a task and inspect the result Use the chapter routes to learn the recorded workflow in stages, with one small project carried through each stage. [HTML](https://lukejbyrne.com/library/guides/9aqh8buatsc/) Content language: en Author: Luke Byrne Video published: 2026-05-11 Guide published: 2026-11-02 [Watch the video](https://www.youtube.com/watch?v=9aQh8buAtSc) Video language: English ## 1. Install and give the first task a concrete home The 11 May 2026 masterclass presents Codex as a workspace for coding and broader file-based work. Installation begins at 3:23 and the overview at 5:17. Start there if you are unfamiliar with the interface, rather than jumping straight to a long-running command. The same small project can serve as a reference while you learn how chats, files and outputs relate to one another. Choose a job whose result you can inspect: a small app change, a file to organise or a clearly described artifact. Give the task a purpose, relevant inputs and a definition of finished. A successful response in chat is only one part of the result; the file or application should demonstrate the requested behaviour. 1. Follow the recorded installation route, then identify the project or folder your first task belongs to. 2. Write the expected result and a simple way to check it before starting the task. 3. At 13:38, explore slash commands and skills, distinguishing reusable instructions from the specific request you are making now. ## 2. Organise projects, inputs and artifacts Plugins and files appear at 17:38, followed by pins, projects and chats at 20:36. Use these chapters to make the working context visible. A connected plugin offers a route to a service, while a file supplies material for the task; neither removes the need to identify the exact input and intended output. Keep related work findable rather than scattering its decisions across unnamed conversations. The artifact section at 24:48 explains another kind of output to inspect. Open the artifact and read or use it in its intended context. For a document, check the content; for an application, complete the relevant task. Keep your comments tied to the version you actually reviewed so a later change does not inherit approval by accident. 1. Give the project and chat meaningful names and pin work you need to find again. 2. Supply the relevant file or connected resource and confirm that the task uses the intended material. 3. Open the produced artifact and compare it with the original request before asking for the next revision. ## 3. Learn automation and parallel work on a bounded example The next route covers natural-language automations at 26:41, image generation at 30:50 and sub-agents at 33:31. These capabilities solve different problems. A reusable instruction is not itself a recurring trigger, an image is an output to review, and parallel agents need distinct responsibilities if their results are to fit together. Begin with one observable run rather than assuming a scheduled task will stay useful indefinitely. For parallel work, decide what each subtask should return and where the combined result will be checked. This makes the work easier to review even when several tasks are progressing at once. 1. For an automation, specify the input, timing and expected output, then inspect an actual run. 2. For image generation, state the intended use and review the generated asset before reusing it elsewhere. 3. For sub-agents, divide a task into bounded parts and review the assembled result, including any overlapping changes. ## 4. Check the work through the browser Browser use starts at 36:27. Follow that chapter when the result lives on a page, because reading a completion message cannot show whether the interface actually works. Choose a short visitor journey that reflects the requested change, and compare the visible outcome with the task’s starting requirements. Keep the check specific. Opening a page, completing a form and confirming saved state are different observations; record only what the task actually demonstrates. If the result is incomplete, describe the visible failure and where it occurred so the next revision has a concrete target. 1. Open the intended page and follow the main task from its starting state to its result. 2. Inspect the visible output and any relevant saved change rather than accepting the agent’s summary alone. 3. Request a targeted repair, then repeat the affected journey after the change. ## 5. Use the CLI chapters to manage a longer session The CLI begins at 41:32. The recording then covers /status at 44:52, /clear at 46:42, permissions and --yolo at 47:40, and /goal at 49:41. Treat these as the recording’s command map: permission changes alter what a session can do and are not a prerequisite for every task. Review the intended scope before changing them. For continuation and housekeeping, use /side at 1:00:53, /compact at 1:02:25, /resume at 1:04:50, /rename at 1:05:44 and /ps at 1:07:26. Keep the current goal, important files and unresolved checks clear when continuing work. A background process or a long-running goal still needs a checked deliverable at the end. 1. Use the usage and session-management chapters to understand the current run before starting another large task. 2. Record what is complete and what remains, then resume the appropriate conversation with that context. 3. Inspect relevant running terminal tasks and the final artifact or application before calling the work finished. ## The takeaway Learn Codex by carrying one task from clear inputs to a reviewed output, then add automation or parallelism where the checked workflow benefits. ## Workflow resources ## Chapters - [0:00 Codex – The Everything App](https://www.youtube.com/watch?v=9aQh8buAtSc&t=0s) - [3:23 Installing Codex](https://www.youtube.com/watch?v=9aQh8buAtSc&t=203s) - [5:17 Codex Overview](https://www.youtube.com/watch?v=9aQh8buAtSc&t=317s) - [13:38 Slash Commands / Skills](https://www.youtube.com/watch?v=9aQh8buAtSc&t=818s) - [17:38 Connecting Plugins + Using Files](https://www.youtube.com/watch?v=9aQh8buAtSc&t=1058s) - [20:36 Organise Projects – Pins / Projects / Chats](https://www.youtube.com/watch?v=9aQh8buAtSc&t=1236s) - [24:48 Understanding Artifacts](https://www.youtube.com/watch?v=9aQh8buAtSc&t=1488s) - [26:41 AI Automations (using Natural Language)](https://www.youtube.com/watch?v=9aQh8buAtSc&t=1601s) - [30:50 Generate Images](https://www.youtube.com/watch?v=9aQh8buAtSc&t=1850s) - [33:31 Sub-Agents (Parallel Tasks)](https://www.youtube.com/watch?v=9aQh8buAtSc&t=2011s) - [36:27 Browser Use](https://www.youtube.com/watch?v=9aQh8buAtSc&t=2187s) - [41:32 Codex CLI](https://www.youtube.com/watch?v=9aQh8buAtSc&t=2492s) - [44:52 /status – Check Usage](https://www.youtube.com/watch?v=9aQh8buAtSc&t=2692s) - [46:42 /clear – Quick New Chat](https://www.youtube.com/watch?v=9aQh8buAtSc&t=2802s) - [47:40 Update Permissions --yolo](https://www.youtube.com/watch?v=9aQh8buAtSc&t=2860s) - [49:41 /goal – for Long-Running Coding Tasks](https://www.youtube.com/watch?v=9aQh8buAtSc&t=2981s) - [1:00:53 /side – Ephemeral Qs](https://www.youtube.com/watch?v=9aQh8buAtSc&t=3653s) - [1:02:25 /compact – Save Tokens](https://www.youtube.com/watch?v=9aQh8buAtSc&t=3745s) - [1:04:50 /resume – Continue where you left off](https://www.youtube.com/watch?v=9aQh8buAtSc&t=3890s) - [1:05:44 /rename – Give chats a memorable name](https://www.youtube.com/watch?v=9aQh8buAtSc&t=3944s) - [1:07:26 /ps – Background terminal tasks (apps running)](https://www.youtube.com/watch?v=9aQh8buAtSc&t=4046s) ## Sources and limitations Prepared 2 November 2026 from this exact published description and chapters. The recording was not independently watched or transcribed and no fresh hands-on test was performed. Steps are editorial adaptations; current setup, availability and offer terms may differ. This is a dated route through the recorded commands, not confirmation of their present availability or plan requirements. The description does not supply exact installation commands. [Original video](https://www.youtube.com/watch?v=9aQh8buAtSc)