How-To Guides

Codex Cheat Sheet: What to Do, in What Order, and Every Skill So Far

A one-page Codex cheat sheet: six steps from install to pull request, each linked to its full article, plus every Codex skill built so far with its GitHub link and the three mistakes to avoid.

  • Six steps in order: install, set up the repo, run a session, plan, build and ship
  • Six skills listed with a one-line description, a GitHub link and the full article
  • ️ Three common mistakes: wrong folder, code before the approach, one session for unrelated tasks
By Rejith Krishnan3 min read
Codex Cheat Sheet: What to Do, in What Order, and Every Skill So Far

The six steps

Each step is a line or two. The link goes to the full article.

1. Install Codex

Pick the guide that matches your machine.

Full articles: WSL install or native PowerShell install.

2. Set up a working repo

Clone the existing Git repo, create a feature branch, and make sure the README and AGENTS.md are in place so Codex has context. The repo-readme-generator skill writes both from the code. For a new project, create a new folder inside your Git repo and start there.

Full articles: repo-readme-generator: README and Docs Skill for Codex and Run Codex as an operating system.

3. Run a session

  • Start Codex from the repo root every time.
  • Use /exit and start a fresh session for each new piece of work.
  • After a PR merges, ask Codex to "pull the latest and switch to main".
  • Install Codex updates whenever it prompts.

Full article: Codex CLI Commands: the essential command line arguments, including YOLO mode.

4. Plan the change

Brainstorm in ChatGPT first. Ask for options, pros and cons, prerequisites and existing repos that already do it, and tell it not to write code. Have it summarize the chosen option, then bring that summary into Codex.

5. Build it

Trigger the plan builder with $plan-builder, answer its questions, switch to plan mode, and run the prompt it produces.

Full article: plan-builder: Ticket-to-Plan-Mode Prompt Skill for Codex.

6. Ship it

Create the PR, add a reviewer, and post the PR link in the Teams chat so progress can be tracked.

Full articles: prepare-pr and review-pr.

Skills built so far

Skill What it does Links
plan-builder Turns a ticket and its artifacts into a complete Plan Mode prompt. Investigates and assesses risk, changes no code. GitHub, article
wave-a11y-scan WAVE-style accessibility scan with findings mapped to WCAG 2.x and a Section 508 note, in a client-ready report. GitHub, article
appsec-static-audit Source-code-only security audit with findings mapped to NIST 800-53, IRS Pub 1075 and EOTSS controls. GitHub, article
repo-readme-generator Creates or updates README.md and the docs suite, including AGENTS.md, from verified repository evidence. GitHub, article
prepare-pr Reviews finished work, updates affected docs, checks the branch name, commits in-scope files and opens the PR after you approve. GitHub, article
review-pr Reviews a GitHub PR against its requirements, scores it on five factors and posts a comment only after you approve the text. GitHub, article

To install every skill in one step, point the installer at the repository root:

$skill-installer install all the skills in the https://github.com/lowtouch-ai/krithi-skills/

Restart Codex afterwards so the new skills are picked up. To install just one, use $skill-installer install with its GitHub link from the table.

Common mistakes

  • Starting Codex from the wrong folder. Always start from the repo root.
  • Asking ChatGPT for code before the approach is settled. Agree the option first, then write.
  • Reusing one session across unrelated tasks. Use /exit and start fresh for each new piece of work.

About the Author

Rejith Krishnan

Rejith Krishnan

Founder and CEO

Rejith Krishnan is the Founder and CEO of lowtouch.ai, a platform dedicated to empowering enterprises with private, no-code AI agents. With expertise in Site Reliability Engineering (SRE), Kubernetes, and AI systems architecture, he is passionate about simplifying the adoption of AI-driven automation to transform business operations.

Rejith specializes in deploying Large Language Models (LLMs) and building intelligent agents that automate workflows, enhance customer experiences, and optimize IT processes, all while ensuring data privacy and security. His mission is to help businesses unlock the full potential of enterprise AI with seamless, scalable, and secure solutions that fit their unique needs.

LinkedIn →