Ask HN: What are you working on? (August 2026)

A friend of mine runs a 6 person company where they specialize in maintaining WordPress sites for SMBs.

It's definitely a viable business!

I built this cli-based project management tracker that is designed to be harness agnostic. Essentially creating a cascading pool of milestones, then sprints, then tasks, each with gates, and context requirements, into a pool for agents to claim. So multiple agents from different harnesses could all work on the same project at once without losing drift of the project's intended focus. Essentially the same project management we all use and love in our day to day. But the cli commands are made for agents while the tui commands are for humans.

https://github.com/mattrichmo/boreal-work/

The thing I love about software, is you occasionally get this holy crap moment and I got that when I had to be away for 2 days and had an orchestrator agent run this for multiple days, without drifting too much. Maybe a little overenginneered but thats whatever. I spent about an hour planning, auditing the plan and adding more so the top-level milestone and inner sprints were pretty well defined or included the relevant context locations for each task. then watching the orchestrator launch a multi-day multi-agent workflow, was quite amazing.

In the immediate, I'm trying to add functionality to my blog, that used to exist before I rewrote it.

It used to be a mashed together set of PHP scripts that used the filesystem for data storage/organization, and I had to write raw HTML for my blog posts. A few years ago I rewrote it in Flask (Python), and designed it to use a database.

One piece of functionality I never got around to copying from the PHP version though, was author biography pages. Early on, my blog was designed such that I could publish things written by other people, because a close friend of mine expressed some interest in having his stuff on it. Thing is, he never got around to writing anything. So when updating to Flask/Jinja, I didn't implement this (I did implement minimal authorship code, so the database already stored author identities, and there was HTML to show author name).

I've been in talks with another friend though about doing the same thing, and I mistakenly told him it was already ready (I thought it was), only to discover the lovely URL `/author/NOTIMPLEMENTED`, so I'm finally creating this endpoint and page template.

I'll also have to update the database schema to include licensing, because the current site always reports the same CC license (the one I chose for all my posts).

https://gitlab.matthewrease.net/matthew/blog

If anyone wants to use the code, there's a Docker image too, and you can easily-ish replace the branding with your own.

Outside of that, for the last few months I've been working with someone on a web app for uploading videos to a server/database.

I'm sick of AI slop, and decided to do something about it.

Even with Fable I'm still seeing wrong but plausible answers, which are the worst kind imo.

AGENTS.md didn't quite give me the right boost either, so here's my flow:

1. Asking 2 new sessions to review what I built in my other session along with guides/standards* 2. Having a third session judge the first two. It dedupes, and does a good job of only surfacing the important stuff 3. Ask it to make edits, etc.

Some missing parts that I've been finding improve review are: - current/past session context - indexed codebase - PRs/git history - research/guides/3rd party context*

I had been running this on my own for a long time, and just started to turn it into a product. It's free right now, would love any feedback (roast me). Looking forward to catching some bugs for you guys.

https://gx.run

Thanks for reading!

* OWASP, Google Code Review, etc

An iOS app that tells you which of your credit cards to use at any store. The part that turned out interesting is getting your targeted bank offers (Amex Offers, Chase Offers etc) without a bank login: a Safari extension reads the offers page you're already signed into in your own browser session, and can tap all the little "add to card" buttons for you. Chase renders theirs inside shadow DOM in a virtualized grid so only the visible tiles exist in the DOM at any moment, which made scraping it way more annoying than it had any right to be. Solo project, about a year in.