First look at Claude Code mods (spoiler: yes, use one now!)

I have to say I was sceptical at first. "Mods" – we already had hooks, right, so what can be so special? I'm not going to retell what mods can do. Nor do I recommend reading through the manual. Instead, ask Claude this:

Claude Code has mods now. Given the workflows we usually use and the type of work we do here, what are some mods we could build that would be super useful?

And let it build some. The key is, they can do everything to your Claude Code and you don't need to know how they work.

These mods and OpenAI's Intelligent UI (released just now) are super interesting steps towards the Disposable Software direction some futurists are talking about. I'm sure Codex, Grok, Gemini and everyone and their mother will have mods very soon.

My first mods

This is how my terminal looks like now, when running a long-running workflow. You will notice the extra right-side pane, that does not look native to Claude Code. And my god, I can resize it with my mouse..

The right side pane opens when a big workflow is running

This mod works in two settings – one for each of the main workflows I described in detail in my previous post – for task preparation and for building.

This is an actual workflow that did task preparation (here only the pane is shown). This specific workflow took two hours to complete and this visualization makes the entire thing understandable with just a glance. Here's an animated sped-up gif of it.

The rectangles are critique rounds where multiple agents criticize the plan. Pink parts are escalation and green is the happy end state

Here's one real recording of the build workflow that writes code and deploys work. This workflow often has tens of discrete steps (like code reviews and fixes, checking something in dev or prod, data heals, merges, deploys etc). This creates a nice overview of it all.

Some dev checks were pink escalations (I had to do the check), also first deploy failed due to a vulnerability scan, and was fixed and re-deployed. Five final checks were done in production by the agent, before calling the run done.

The second thing I could fix now, was my long time feature request, to always have a hyperlink where Claude refers a Github issue. Of course Claude knows what #2443 (and three other ticket references it casually mentions) means, because it has the context, but I don't.

So now I have a mod for this.

Make it so that whenever I see a ticket number in the terminal, it is a hyperlink

Amazing!

Next mods for me

These are all suggestions by Claude, which I love a lot. I'll get to them soon, but I'm listing them here to share some inspiration.

Jira-like task board in terminal, so I can skip going to Github issues
Deploy watcher and status light
Move a lot of prose and rules from agents.md into deterministic checks
Safer tool for querying production data, no hand-assembling the SQL proxy
A queue and lock for running parallel e2e tests. I really love this! Today, my parallel agent count is limited by how many agents can run the memory-intensive browser-based tests simultaneously. They rarely actually end up doing it at the same time, but limit needs to be there still. This mod would act as a queue with locks that would only start limiting agent counts when they actually want to run the tests at the same time.

Someone made a Jev-based model switcher. Somebody made a mod where you can play Doom in a terminal pane while you wait for the model.