Scott Spence's Blog

I’ve upgraded Coolify twice this week, on two different servers, and both times the dashboard was unusable afterwards.

Back in March I wrote about enabling LSP in Claude Code.

In June I wrote about how I stop LLMs drifting in production codebases.

Claude Code keeps your sessions on disk, for 30 days by default.

I’ve had syntax highlighting on this blog since I moved it over to mdsvex back in 2021.

On a reinsurance platform I built at an agency this year, the documentation got big.

Since January 2025 I’ve built 21 MCP servers and tools.

I created a development setup from EndeavourOS to Ubuntu 26.04.

September 2023 is when I started using AI as part of my job.

Agentic engineering is the practice of putting coding agents inside an engineering system that makes context, scope, validation, evidence and human review explicit.

In November 2025 I wrote the honest version of working with Claude Code.

I have a habit of turning repeated arguments with coding agents into software.

The problem I care about with LLM coding is drift: one plausible shortcut gets copied, then future sessions start treating it as how the app works.

Right, so, this started with me looking at context-mode, seeing a 98% context reduction claim, and immediately thinking: bull shite.

I have written a few posts now about bits of my-pi: adding LSP, adding telemetry, and hardening redaction with evals.

Third post in a little run about my-pi, my coding agent harness built on top of pi-coding-agent.

my-pi keeps growing in the direction of whatever annoyed me most that week.

This whole thing started with a chat with Kev, (Svelte Society co-founder), after a few beers at the CityJS after-party.

So, here’s the situation.

I’ve been using the UserPromptSubmit forced-eval hook for months now.

I saw a Reddit post the other day that called setting up MCP servers in Claude Code “a tech ritual for the quietly desperate.” 😅 Keep reading

Right, so, I stumbled across this post by Karan Bansal about enabling LSP in Claude Code and thought “no way this actually works.”

Ok, so, I’ve built over 20 MCP servers at this point.

Remember that post where I tested skill activation with the Claude API and got the forced-eval hook to 84%?

Right, so, remember when I wrote about unlocking swarm mode with claude-sneakpeek?

Ok, so, I’ve been using the Claude Code swarm feature now for several days and I’m super impressed!

So, I watched a video about Kimi K2.5 from Sam Witteveen that detailed a load of cool looking features but what caught my eye was the swarm mode feature, this is where it will spin up to (apparently) 100 sub-agents to work on tasks in parallel.

Aight, so, continuing with the Claude Code ‘bits’ and my toolchain I’m putting together for working with Claude Code.

This is the follow-up to Local Analytics with SQLite and SvelteKit.

Real quick on this one!

I’ve been using Fathom Analytics for years and it’s great, but I wanted something more integrated.

I spent some time over the last few days evaluating some Linux distros, two to be precise.

So, remember that blog post I wrote about Claude Code skills being a coin flip?

So, I’m a team lead working on a large monorepo for a private client.

So I got Claude Skills to register with Claude Code (sorted the YAML formatting issue), and the skills are now showing up when I ask Claude to list them.

Not having Claude Code skills showing up?

I was looking up something on my site for reference (as I usually do when trying to recall a detail) when I noticed the popular posts in the footer weren’t displaying.

So there I was, starting a Claude Code session, and I noticed something disturbing and completely of my own making: my MCP tools were consuming 66,000+ tokens of context before I even started a conversation.

I spent some time this weekend rolling the auth credentials for this site (again) after Claude Code doxed my .env variables!

I’m currently in the process of migrating a large SvelteKit monorepo from Svelte 4 to Svelte 5 with my team.

Cool!

In this post I’m going to go through converting the SvelteKit minimal template from using @testing-library/svelte and jsdom over to using @vitest/browser, vitest-browser-svelte and playwright.

Keep reading

Ok, off of the back of an absolutely banger of a post from Ilja, I’m now compelled to write about my own experience at Svelte Summit.

I poisoned the well!

So, since the Model Context Protocol (MCP) was announced at the end of November 2024 I immediately saw the potential that this could offer me.

Aight!

I used Windsurf for the whole day yesterday, I’m impressed!

Aight!

So, this is my updated Zsh config!