Open source and developer tools, tested before we write about them
RepoSquare answers the questions developers ask before they install something. Is this repo worth it? How does it compare with the tool I already use? Why is it not doing what the README promised? Each comparison and guide comes from running the tool on a small real task. We note the version, the commands and what broke, and we update posts when tools change. New here? Start with one of the hands-on posts below, or read how RepoSquare works.
Start here
- Busbar vs Bifrost vs LiteLLM: Open Source LLM Gateways Tested: three gateways on the same small server, with the main AI provider broken on purpose to time the failover.
- Claude Code Not Reading AGENTS.md? 19 Tests, 4 Fixes: why Claude Code silently skips AGENTS.md when a CLAUDE.md exists, and the one-line fix.
- How Do I Make Claude Code Websites Not Look AI Generated?: 14 pages built with four design skills to see which one actually changes the look.
- xan vs csvkit: Fast CSV Stats and Filters in the Terminal: both tools run on the same 200,000-row file, with time and memory measured.
- Can an Old Android Phone Be a Real Linux Server Without Root?: ServerBox vs Podroid, what PRoot can and cannot run, and whether Claude Code works on it.
- ExifCleaner vs ExifTool: how to strip GPS and camera data from photos, and how to check that the file is actually clean.
Latest posts
- Can Claude Code Clone an App?
You see an app you like and think, “I could build that.” Then you open Claude Code, paste the link, and type “clone this.” Does it work? And… - motion2code vs HyperFrames
A coding agent (a program that edits your files and runs commands, not a chat box that only talks) can write a pretty animation and still be wrong… - ExifCleaner vs ExifTool
I dropped a photo of a desk into a chat and treated the file as the picture. The picture was only the colored squares. Beside those squares the… - How Do I Benchmark My VPS?
Here is how to benchmark a VPS. Test four things: CPU, memory, disk and network. Use sysbench for CPU and memory, fio with direct I/O for disk, and… - Updating WordPress does not close PHP2Shell
PHP2Shell is two use-after-free bugs in PHP itself, not in WordPress, a plugin or a theme. Which PHP versions fix it, why the 70 percent headline is a survey, why the php.ini settings are only partial, and what to check before lunch. - What Is a Lightweight Alternative to Electron for Python?
Short answer: the best Python Electron alternative depends on what your app draws. If you want a web-style UI (an interface built with HTML and CSS), pywebview is… - How do I push a Docker image without a registry?
You can push a local Docker image to your own server without a registry: docker pussh sends only the layers the server is missing over SSH, then removes its temporary shelf. Install steps, flags, and when a real registry still wins. - How Do I Free Up Disk Space From Claude Code?
Claude Code keeps every session in a hidden ~/.claude folder. It already deletes most old files after 30 days, cleanupPeriodDays shortens that, claude purge wipes one project, and one popular cleaner deletes your plugins on its defaults. - Huashu Motion vs CapCut
Huashu Motion draws a product film from one HTML page when you have no footage; CapCut cuts the footage you already shot. Here is how the om CLI works, what om doctor needs, and how transparent titles land back on a CapCut timeline. - Pele vs Mermaid
Pele draws the same Mermaid text as Mermaid, 127 to 217 times faster on four timed diagrams, from a browser file about six times smaller. Here is how to use it, what it draws, and when one chart is better left on Mermaid. - Four Coding Agents Beat Minecraft. They Never Saw the Screen.
Google DeepMind’s four Gemini 4 Argon agents beat the Ender Dragon in 27h 14m through Mineflayer text, not the screen. The setup, the step timeline, a recomputed cost and stronghold check, the six limitations, and what to copy. - Which Claude Code Plugins Are Worth Installing?
Most official Claude Code plugins are cheap to keep on. All 39 measured one at a time: fourteen add zero tokens, three big kits add 2,000 to 3,100 per request, the built-in estimate misses hook text, and security-guidance runs a paid Opus review.