My local tmux and one remote tmux both used Ctrl-k as their prefix. After connecting over SSH, Ctrl-k c still created a local window. The local tmux consumed the prefix before SSH could send it to the server.

I wanted Ctrl-k to control the remote when its tmux also accepted Ctrl-k, but stay local for other servers. The working solution uses a pane-local flag and one SSH probe per login. Ctrl-b remains an always-local escape hatch.

My Linux Mint notebook accepted SSH connections while its lid was open, but became unreachable when I shut it. The connection command was unchanged:

1

ssh william@127.0.0.1 -p 62223

The relevant setting was not in sshd_config. Cinnamon was configured to suspend the notebook when the lid closed, even on AC power. I changed just the AC lid action, leaving the battery policies alone.

I wanted Claude Code’s terminal interface and tools to run ChatGPT models. Installing gptcc was only the first step. I could select Sol in /model and still get an account-hold error before the request reached GPT.

The working setup uses a local authenticated gateway, ChatGPT OAuth, and Claude Code’s custom model picker. It lists eight GPT models, makes Default resolve to GPT-6.1 Sol with medium effort, and displays Codex’s five-hour and weekly usage beside the model. A real Claude Code session completed a file-read tool call in two turns.

I wanted to know which tasks in trading.hyena17 make the morning setup run past 09:30. Sorting by duration was misleading. Some tasks deliberately run until 15:00, while a task that takes only five seconds can still start ninety seconds late.

The Nov 12 run has two main problems. Stock-weight preparation takes 352.6 seconds and leaves only 4.4 seconds before 09:30. Downstream, the slowest shared-service startup takes another 167.3 seconds and finishes at 09:35:15. Redis itself takes 4.8 seconds. Its late start comes from its dependencies.

Update 2026-11-11 17:30: image tags are now the build time, yyyymmddTHHMMSS (e.g. airflow:20261111T172041), so two builds on the same day no longer collide, and ./run.sh with no tag lists what’s available. The outputs below are from the first build, which still used a plain date tag, 20261111. The downloadable scripts are the current versions.

The outage postmortem had one recurring theme: every time someone rebuilt the Airflow container by hand, it came back missing something. A mount, the concurrency settings, the plugin, the registry login. The image was whatever someone last typed, and the settings that mattered lived only in a running container.

So the same afternoon I wrote three small scripts. build.airflow.sh turns the running container into an image tagged with its build time (airflow:20261111T172041), pushes it, and saves a tarball. run.sh recreates the container from that tag and checks it’s actually healthy, or puts the old one back. restart.sh runs inside the container and keeps the scheduler and webserver up. From the 170 entries in docker diff, the image keeps 13 files (172K). A redeploy is healthy about 15 s after the new container starts, and a failed one rolls back in 1 s.

Two pages I look at every day had the same problem: monospace text that was too thin and pale to read comfortably. One was the log viewer in our Airflow 2.7 web UI, the other the dark theme of an internal Tabler dashboard. In both, Chrome was drawing the text in Courier New, a hairline font, even though both pages list four or five better fonts before it.

The cause was a fontconfig file from the fonts-powerline package, installed in 2019. It puts the symbols-only “PowerlineSymbols” font in front of most monospace families, and Chrome then refuses every one of them. A one-rule fontconfig file fixed it.