rlnorthcutt

Author here. The thing that surprised me writing this was how wrong I understood the AOT story initially. I'd repeated the "native binary, no PHP needed" line, but thats not really the case: binary mode still links libphp and PHPX, which have to ship in your deployment package. It starts without the PHP CLI, but it isn't a static Go-style binary.

The more interesting mode is -m ext, which outputs a .so that PHP loads as a normal extension. Writing a PHP extension has always meant writing C. Being able to write one in PHP changes who can realistically do it.

So, now PHP devs can ship better CLI tools (with libraries, but with less environment issues), AND can create custom PHP extensions that can speed up parts of the app.

"There are moments when a dependency graph becomes performance art."

Hats off to the author for not only following the muse into the archives of ancient tech... but for stubbornly continuing forward even when all paths seem open.

I loved the trip down memory lane with some of these things, and really resonate with the dependency hell issue... though the hardware nature of this and multiple machines is a nice twist.

Really fun read. 5 Stars. Would read again.

Omnideck stress-tests its own browser tools on 50 real websites, diagnose failures by reading its own own source code, and files GitHub issues with the fix already identified.

It does NOT apply the fix itself to maintain separation of concerns and get human in the loop quality checks and planning.

A practical look at the two OpenAI API formats, why one became a de facto standard, and what the new Open Responses spec changes.

I built analyzerepo to solve the "cold start" problem for both new developers and AI agents. Whether you are jumping into a legacy codebase or trying to make Claude Code actually useful on day one, you usually spend the first hour just trying to figure out where the entry points are and how the pieces fit together.

This is a Go-based CLI that points Claude at any GitHub or local repo to generate three specific, high-context Markdown files:

1) ONBOARDING.md: A human-readable guide that maps out the project’s purpose, language stats, file tree, and a "doc library" of existing .md files. Hand this to a new contributor to save an hour of verbal walkthroughs.

2) ANALYSIS.md: A per-file audit that classifies roles (e.g., entrypoint, core, util) and provides structured improvement suggestions. Each suggestion includes a done_when condition specifically optimized so you can paste the block directly into a Claude Code prompt for a working implementation.

3) CLAUDE.md: A project-specific context file that Claude Code reads automatically to understand your architecture and conventions instead of being generically cautious.

Key Features:

- Zero Dependencies: No runtime or language install required; it’s a single binary. - Smart Selection: It uses Claude to identify the most structurally significant files instead of hitting token limits with junk files. - Flexible Backend: It automatically detects your ANTHROPIC_API_KEY or uses your existing Claude CLI installation.

I’d love to hear how this handles your repos or what other "AI-native" documentation formats would be useful for your workflow.

Wow - I really appreciate you taking the time to look at it again. My original comment was written quickly, and probably no where near as clear as it could have been.

I respect your willingness to modify your original stance upon closer examination. Non-ironic hat tip.