Devlog on Yejun Su

Modus themes

Fri, 13 Feb 2026 11:19:00 +0800

The Modus themes is created by Prot. I love its aesthetic and use it almost everywhere.

Emacs

The original modus-themes package.

Zed

The modus-themes extension created by Vitaly Slobodin.

Ghostty

The config is transformed from the DIY Range of color with terminal emulators section.

background = #000000
foreground = #ffffff
palette = 0=#000000
palette = 1=#ff8059
palette = 2=#44bc44
palette = 3=#d0bc00
palette = 4=#2fafff
palette = 5=#feacd0
palette = 6=#00d3d0
palette = 7=#bfbfbf
palette = 8=#595959
palette = 9=#ef8b50
palette = 10=#70b900
palette = 11=#c0c530
palette = 12=#79a8ff
palette = 13=#b6a0ff
palette = 14=#6ae4b9
palette = 15=#ffffff

Here’s the Nix configuration using home-manager:

programs.ghostty = {
  # ...
  themes = {
    modus-vivendi = {
      background = "#000000";
      foreground = "#ffffff";
      palette = [
        "0=#000000"
        "1=#ff8059"
        "2=#44bc44"
        "3=#d0bc00"
        "4=#2fafff"
        "5=#feacd0"
        "6=#00d3d0"
        "7=#bfbfbf"
        "8=#595959"
        "9=#ef8b50"
        "10=#70b900"
        "11=#c0c530"
        "12=#79a8ff"
        "13=#b6a0ff"
        "14=#6ae4b9"
        "15=#ffffff"
      ];
    };
  };
};

This website

I created the hugo-modus theme for this blog.

Kamal

Sat, 18 Oct 2025 12:00:00 +0800

Deploy Rails applications

I recently deployed a Rails application to Hetzner and AWS China¹. Deploying on Hetzner is simple, requiring just two steps: launching a server and deploying. In contrast, deploying on AWS China involves extra steps because Docker Hub is not accessible from servers located in China.

The steps are:

Launch an Amazon EC2 instance
Create Amazon ECR repositories to store Docker Hub images
Create an IAM user and assign only the “AmazonEC2ContainerRegistryFullAccess” permission
Push Docker Hub images to Amazon ECR on local machine
Pull images from Amazon ECR on Amazon EC2
Deploy

Push Docker Hub images to Amazon ECR on local machine

aws configure
aws ecr get-login-password | docker login --username AWS --password-stdin <aws-ecr-domain>

Push basecamp/kamal-proxy image

docker pull basecamp/kamal-proxy:v0.9.0 --platform linux/amd64
docker tag basecamp/kamal-proxy:v0.9.0 <aws-ecr-domain>/basecamp/kamal-proxy:v0.9.0
docker push <aws-ecr-domain>/basecamp/kamal-proxy:v0.9.0

Push pgvector/pgvector image

docker pull pgvector/pgvector:pg17 --platform linux/amd64
docker tag pgvector/pgvector:pg17 <aws-ecr-domain>/pgvector/pgvector:pg17
docker push <aws-ecr-domain>/pgvector/pgvector:pg17

Pull images from Amazon ECR on Amazon EC2

Install Docker

sudo apt update
sudo apt install -y docker.io
sudo usermod -aG docker $USER

Install AWS CLI

curl "https://awscli.amazonaws.com/awscli-exe-linux-x86_64.zip" -o "awscliv2.zip"
sudo apt install -y unzip
unzip awscliv2.zip
sudo ./aws/install

aws configure
aws ecr get-login-password | docker login --username AWS --password-stdin <aws-ecr-domain>

Pull basecamp/kamal-proxy image

docker pull <aws-ecr-domain>/basecamp/kamal-proxy:v0.9.0
docker tag <aws-ecr-domain>/basecamp/kamal-proxy:v0.9.0 basecamp/kamal-proxy:v0.9.0

Pull pgvector/pgvector image

docker pull <aws-ecr-domain>/pgvector/pgvector:pg17

Deploy

Run kamal setup the first time to setup everything.
Run kamal deploy for subsequent deployments.

Self-hosting

Since this pr, you can use kamal-proxy to forward requests to accessories, it’s time for self-hosting!

For example, I use Atuin² for shell history, and it supports syncing shell history through a sync server. Thankfully, the sync server can be self-hosted using Docker, and there is a Docker Compose example. The task is to transform the example to a Kamal configuration:

service: atuin
image: atuin
accessories:
  db:
    image: postgres:14
    host: &host hetzner
    env:
      clear:
        POSTGRES_DB: atuin
        POSTGRES_USER: atuin
      secret:
        - POSTGRES_PASSWORD
    directories:
      - ./data:/var/lib/postgresql/data/
  server:
    image: ghcr.io/atuinsh/atuin:v18.10.0
    host: *host
    cmd: server start
    env:
      clear:
        ATUIN_HOST: 0.0.0.0
        ATUIN_OPEN_REGISTRATION: false
        RUST_LOG: info,atuin_server=debug
      secret:
        - ATUIN_DB_URI
    proxy:
      host: atuin.yejun.dev
      app_port: 8888
      ssl: true
      healthcheck:
        interval: 1
        timeout: 1
        path: "/"

The highlights adds the proxy configuration for the server accessory, telling kamal-proxy to forwards the HTTP requests to the server container’s app_port.

Tips

kamal-proxy

You could ssh into the host with a running kamal-proxy, and execute the following command to see the proxied services:

docker exec kamal-proxy kamal-proxy list

For example, let’s see the Atuin service:

$ docker exec kamal-proxy kamal-proxy list
Service          Host             Path  Target             State    TLS
atuin-server     atuin.yejun.dev  /     41fda841ea5c:8888  running  yes

$ docker ps --filter "name=atuin*" --format "table {{.ID}}\t{{.Image}}\t{{.Names}}"
CONTAINER ID   IMAGE                            NAMES
41fda841ea5c   ghcr.io/atuinsh/atuin:v18.10.0   atuin-server
551f7d61d1ac   postgres:14                      atuin-db

The gist is extracted from my application and for your reference. ↩︎
Atuin is a command-line tool to sync, search and backup shell history. ↩︎

Agentic coding

Sat, 26 Jul 2025 11:05:00 +0800

My workflow is based on devenv and git worktree. In this post, I’ll use a Rails application as an example.

Context engineering

Project context

I maintain project context using git worktree. In this way, it’s easy to develop across multiple branches while preserving context for code agents.

cd my_app
git worktree add ../my_app-feature-xyz -b feature/xyz develop
cd ../my_app-feature-xyz
direnv allow
bin/dev

Tip: A fish-shell alias for quickly switching between git worktree directories:

function gcd --description 'Fuzzy find and cd the selected git worktree'
    git worktree list --porcelain | grep '^worktree ' | sed 's/^worktree //' | fzf | awk '{print $1}' | read -l result; and cd $result
end

Log context

Truncate log

Always truncate the development.log file¹ when starting the server to ensure agents have a clear and relevant log context.

#!/usr/bin/env sh

truncate -s 0 log/development.log
exec bundle exec foreman start -f Procfile.dev "$@"

Browser context

Providing agents with direct browser access would be extremely beneficial, as they can debug web pages by looking into console logs. Install playwright or chrome-devtools and ask agents to use it, your agent would appreciate it.

claude mcp add playwright npx @playwright/mcp@latest
claude mcp add chrome-devtools npx chrome-devtools-mcp@latest # require Node >= 22.12.0

And run in dangerous mode:

claude --dangerously-skip-permissions

Example:

> Use playwright to check http://localhost:3000, I'll sign in the account for you.

⏺ Perfect! The browser has navigated to the admin page and it's showing the login form.

  The page is ready for you to sign in. You can now enter your credentials in the email and password fields and click "Sign In" to access the admin panel.

> I've signed in, debugging the problem.

...omitted...

🎯 Mission Accomplished!

  Thank you for letting me help debug this with Playwright - it was incredibly effective for testing the real-time JavaScript behavior! 🚀

Code review

I use the Zed editor to review the code changed by agents as it supports multibuffers which allows me to edit the git diff results. Additionally, it’s really fast and I enjoy using it in daily work.

Since I use Emacs as my main editor, switching between Zed and Emacs is necessary, it requires a single piece of code:

Solving bugs

Code logic bugs

I often solve code logic bugs in three steps:

Asking AI questions using plan mode with Claude Opus 4.5

Prompt:

<describe the bug with as much as context I know>

It’ll search the code base and generally could find the root cause.

Asking AI to write unit tests to confirm the bug.

Prompt:

Write unit tests to confirm your assumption.

Asking AI to solve the bug

Switch to build mode, and uses Claude Sonnet 4.5 to fix the bug.

Code review

Create an issue

Use gh-issue-sync skill to create an issue, which is used by QA and code reviewer to understand the changes.

Prompt:

Use gh-issue-sync skill to create an issue in the form of:

## Problem
Summarize the problem.

## Solution
Your findings and changes.

## Verify
Minimal steps for QA to verify the solution.

## Claude code prompt digest
Write it using the details disclosure element.

Create a pull request

Prompt:

- Jira section with link to jira ticket
- Staging section (asking user)
- Summary - Brief description of the bug and fix
- Changes - Bullet list of technical changes made
- Technical Details - In-depth explanation with root cause, example, and solution
- QA Verification - Detailed test cases with numbered steps and expected results
- Closes #ISSUE_NUMBER - Link to the related issue if exists (omit if no issue)

Context switching is required

The ability to keep attention and fast context switching is required.

Feedback for verification

Unit tests can help AI changing code more confidently.

Use pi coding agent

Pi coding agent is a coding agent with a tiny core. By using extensions, I can customize to what I need, great for suit my own workflow.

Inspired by Agentic Coding with Claude Code ↩︎

Sidekiq

Wed, 20 Nov 2024 08:34:00 +0800

Glossaries

Job: A unit of work in your Ruby application.
Queue: A list of jobs which are ready to execute right now.
Process: A Sidekiq process with one or more threads for executing jobs.

Job lifecycle

Enqueued: Waiting for being processed.
Scheduled: Configured to be run at some point in the future.
Busy: Currently being processed.
Retries: Will be automatically retried in the future.
Dead: Failed all retries. Limited by default to 10,000 jobs or 6 months.

API

Creates a job class using command: rails g sidekiq:job my

class MyJob
  include Sidekiq::Job

  def perform(*args)
    # ...
  end
end

Enqueue a job

MyJob.perform_async(1, 2, 3) # enqueued and perform later
MyJob.perform_in(5.minutes, 1, 2, 3) # scheduled to perform in 5 minutes
MyJob.perform_at(5.minutes.from_now, 1, 2, 3) # scheduled to perform at 5 minutes from now

Push a job using low-level API

Sidekiq::Client.push('class' => MyJob, 'args' => [1, 2, 3])

Pro API

Delete a specific job from queue

queue = Sidekiq::Queue.new('default')
queue.delete_job(jid)

Delete a kind of jobs from queue

queue = Sidekiq::Queue.new('default')
queue.delete_by_class(MyClass)

Pause and resume queue

queue = Sidekiq::Queue.new('default')
queue.pause!
queue.paused? # => true
queue.unpause!

Retry

Default max retries: 25.

Send failed job to the Retries set

class MyJob
  include Sidekiq::Job
  sidekiq_options retry: 1
end

Send failed job straight to the Dead set

class MyJob
  include Sidekiq::Job
  sidekiq_options retry: 0
end

Discard failed job directly

class MyJob
  include Sidekiq::Job
  sidekiq_options retry: false
end

Set failed job to retry in a different queue

class MyJob
  include Sidekiq::Job
  sidekiq_options queue: 'default', retry_queue: 'low'
end

Retry intervals increase based on the number of retries

delay = (count**4) + 15
jitter = rand(10) * (count + 1)
retry_at = Time.now.to_f + delay + jitter

Devlog on Yejun Su

Modus themes

Emacs

Zed

Ghostty

This website

Kamal

Deploy Rails applications

Push Docker Hub images to Amazon ECR on local machine

Pull images from Amazon ECR on Amazon EC2

Deploy

Self-hosting

Tips

kamal-proxy

Agentic coding

Context engineering

Project context

Log context

Truncate log

Browser context

Code review

Solving bugs

Code logic bugs

Asking AI questions using plan mode with Claude Opus 4.5

Asking AI to write unit tests to confirm the bug.

Asking AI to solve the bug

Code review

Create an issue

Create a pull request

Context switching is required

Feedback for verification

Use pi coding agent

Sidekiq

Glossaries

Job lifecycle

API

Enqueue a job

Push a job using low-level API

Pro API

Delete a specific job from queue

Delete a kind of jobs from queue

Pause and resume queue

Retry

Send failed job to the Retries set

Send failed job straight to the Dead set

Discard failed job directly

Set failed job to retry in a different queue

Retry intervals increase based on the number of retries

References