This page is dedicated to relevant snippets/plugins/topics that I have used or am using for work related purposes.
News articles
Code libraries/add-ons
General
Backlog.md is an agent based backlog manager which is aimed at stabilizing and improving desired outcomes of prompts by creating tasks with certain requirements and plans in place, constantly reminding the agent to stay within that scope.
Backlog.md
MrLesk • Updated Nov 7, 2025
MCP related
Context7 - Up-to-date documentation for LLMs and AI code editors
Generate context with up-to-date documentation for LLMs and AI code editors
Guides: Next.js MCP Server
Learn how to use Next.js MCP support to allow coding agents access to your application state
Workflows
Please, use git.
Codex Web (ChatGPT Plus only)
If you want something straightforward with almost no setup, this is the way I go. (and went at first)
You go to chatgpt.com/codex
How it works
- Codex runs in the browser, but executes commands in a managed sandbox on OpenAI’s side.
- You can mount a GitHub repo, edit files, run builds, deploy — all without touching your local machine.
- Approvals are web-based: Codex will ask before writing files, committing, or pushing.
Pros / Cons
Pros
- Zero setup: works anywhere you have a browser.
- No Docker Desktop, no WSL, no local dependencies.
- Safer than YOLO mode: actions require explicit web approvals.
Cons
- Slower feedback loop: commands run remotely, not on your machine.
- Limited integrations: less control over caching, networking, local tooling.
- Still a security risk — you’re pasting commands into a remote shell.
When to use
Codex Web is great for lightweight work: quick edits, config fixes, or experimenting in a clean sandbox. For deeper local integration (fast builds, custom dev stacks), Docker or Compose is still the better option.
Codex in Visual Studio CLI (Open Source possible)
If you already live in VS Code or Visual Studio, you don’t need Docker or the web interface. You can run Codex directly in your project terminal.
Open a bash shell in the editor and start Codex with:
Breaking down the command
codex→ the CLI client.
-ask-for-approval never→ disables approval prompts. Codex won’t stop to ask before editing or running; it just does it. (⚠️ In my experience, this didn’t work)
-sandbox workspace-write→ gives Codex write access to your project workspace. Everything else is read-only.
In other words: no prompts, direct write access to your codebase.
This way you will have to give it approval for edit’s it makes to your code, and depending on your
config.toml and your version of codex, you might also need to approve commands it runs.Pros / Cons
Pros
- Seamless: Codex runs right where you code, in the same terminal.
- Fast loop: no GUI clicks, no Docker overhead.
- Approvals disabled = maximum convenience. (if it works for you)
Cons
- Security risk: Codex can modify anything in your workspace.
- Less guardrails: if it makes a bad edit, it’s immediately in your repo.
- Easy to get lazy and stop reviewing changes.
- If you’re on a windows machine, you need WSL2
- If you’re using WSL2, it seems to not be able to run all linux commands, so it has to use other commands. This eats up a big part of it’s context.
Codex Full YOLO Docker CLI (Open Source possible)
Sometimes you just want to run Codex with zero friction. No approvals, no context switching, no “are you sure?” prompts. That’s full YOLO mode.
The idea:
- Have a Github repo + tech stack + hosting pipeline that automatically builds and deploys on push.
- Run Codex inside a container that mounts your project and Codex state.
- Approve once using(
/approvals), then give it full write access.
Running it directly
With Docker Compose
Pros / Cons
Pros
- Runs inside Linux: low context overhead, feels native.
- Fast startup, no manual approvals needed.
- Convenient for experimentation or continuous development.
Cons
- Security risk: if Codex executes malicious code, it has broad access.
- Encourages laziness: skipping reviews means skipping safeguards.
⚠️ YOLO means YOLO: fast and fun, but you’re trading away safety nets.
Use cases
Use codex for an ai powered changelog writer
Run Codex head-less in pipelines. Example GitHub Action step:
name: Update changelog via Codex run: | npm install -g @openai/codex export OPENAI_API_KEY="${{ secrets.OPENAI_KEY }}" codex exec --full-auto "update CHANGELOG for next release"