public digest · 5 picks
Modern C++ to market bots: five repos worth your time
Five picks this week that cover a lot of ground — a decade of C++ evolution compressed into one readable tutorial, the Ruby coverage tool that's been quietly getting better while you weren't looking, and a Skia-backed canvas library that actually ships working binaries. Also in: an opinionated workflow layer for teams running Codex CLI at scale, and a hackathon-born trading bot with genuinely interesting ideas alongside some things you should know before pointing it at real money.
As always, we include things we think are worth your time, not things that are merely popular. A few of this week's picks have sharp edges worth knowing about before you commit — we've called them out directly.
// pick 1 of 5
A book-style tutorial covering C++ from C++11 through C++26, aimed at developers who already know traditional C++ and want to get up to speed on the modern features fast. Available as GitHub markdown, PDF, EPUB, and a Hexo-powered website. Written by a single author, originally in Chinese, with an English translation.
This is the C++ tutorial you reach for when you already know the language and need to close the gap between what you learned and what the standard looks like now. The C++11 through C++17 chapters are genuinely dense in the good sense — each feature gets historical context explaining why the committee added it, which makes the rules stick better than syntax-only references. Runnable examples are organized by chapter number so you can compile as you read rather than copy-pasting from a webpage.
The C++23 and C++26 chapters are noticeably thinner than the earlier material — treat them as a pointer to what exists rather than a complete reference. The English translation is careful but shows its seams; technical points occasionally land slightly off. One hard practical note: the CC BY-NC-ND license prohibits modified redistribution, so if you were planning to fork this for internal training docs, you can't.
View on GitHub → Our full take →
// pick 2 of 5
oh-my-codex (OMX) is a workflow layer that wraps OpenAI Codex CLI with structured planning/execution phases, team coordination via tmux worktrees, persistent state in `.omx/`, and a catalog of reusable skills and role keywords. It is for teams already using Codex CLI who want a repeatable agentic workflow without wiring it up themselves. If you are not already committed to Codex CLI as your AI coding engine, this is not the entry point.
If your team is already running Codex CLI and finding that vanilla agentic sessions devolve into agents overwriting each other's work, the worktree isolation model here is the right answer. Each session gets its own git worktree, and the planning pipeline ($deep-interview → $ralplan → $ultragoal) gives you a concrete handoff point between spec and execution that most teams are currently making up ad hoc. The Rust runtime core — authority, dispatch, mailbox, replay — suggests real thought went into the state machine rather than shelling out to bash between prompts.
The conceptual surface area is enormous, though. HUDs, missions, sparkshell, wiki, MCP bridges, OpenClaw, hermes-bridge, herdr-bridge — this is accumulating features faster than they're stabilizing, and Windows support is explicitly second-class. The bigger structural risk: the entire thing couples tightly to Codex CLI internals, so an API change upstream can break things without warning. Worth tracking closely, but probably not the production bedrock for a team that can't absorb that kind of churn.
View on GitHub → Our full take →
// pick 3 of 5
SimpleCov is the standard code coverage tool for Ruby, wrapping Ruby's built-in Coverage library with a much more usable API for filtering, grouping, merging results across test suites, and generating HTML reports. It's been the default choice in Rails projects for over a decade and has grown to support branch, method, and eval coverage in addition to the original line tracking. If you're writing Ruby tests and want to know what you're missing, this is the tool.
SimpleCov has been the default Ruby coverage tool long enough that most people installed it once and stopped thinking about it. That's a shame, because the branch coverage support added in recent versions is the feature that actually tells you something. Line coverage on a one-liner conditional is nearly useless; the branch annotation in the HTML report — showing hit counts per arm — makes the missed paths obvious in a way line numbers never could.
The multi-suite merging story is also solid: parallel CI workers accumulate results into `.resultset.json`, and the `SimpleCov.collate` path gives you a clean separation between per-worker collection and final reporting. A few things to know going in: the `# simplecov:disable` directive deprecating `:nocov:` will generate warnings across any project that standardized on the old syntax, and the implicit-else branch noise from `||=` and `case/when` will inflate your missed-branch counts until you find the `ignore_branches :implicit_else` option. The HTML formatter now has a full Bun/TypeScript build pipeline — fine for the maintainers, but non-trivial if you want to extend it.
View on GitHub → Our full take →
// pick 4 of 5
A Node.js native addon that wraps Google's Skia rendering engine via napi-rs, giving you a drop-in Canvas 2D API with no system dependencies. It ships prebuilt binaries for every major platform, so you install it and it works — no libcairo, no ImageMagick, no Pango. The target audience is Node.js backends that need server-side image generation: social card generators, chart renderers, PDF thumbnails.
The zero-system-dependency claim is the whole pitch, and it holds. The prebuilt binaries bundle everything — no libcairo install, no Pango configuration, no mystery build failure on a fresh Docker image. For server-side image generation in Node.js (social cards, chart exports, PDF thumbnails), this is currently the cleanest path. Performance comes in measurably faster than node-canvas on gradient workloads, and the Path2D boolean operations are a genuine extension beyond the browser spec that SVG tooling will actually use.
A few heads-ups: Lambda support depends on a third-party layer not maintained by this project, so if that repo goes quiet you'll need to handle it yourself. The glibc 2.18 minimum will silently fail on certain Alpine images with an error that isn't immediately obvious. Building from source is legitimately painful — Skia is a submodule and the documented setup involves sudo-installing clang as a one-liner. If you're consuming prebuilt binaries you'll never see this, but it's worth knowing if you ever need to build for an unsupported target.
View on GitHub → Our full take →
// pick 5 of 5
CloddsBot is a self-hosted, chat-driven trading terminal that routes natural-language commands through Claude to execute on prediction markets (Polymarket, Kalshi), crypto spot/perps, and Solana/EVM DeFi. It targets developers who want a single bot managing multiple trading venues without building each integration themselves. The catch: it was built in 12 days for a hackathon, and it handles real money.
The integration surface here is the genuinely interesting part: prediction markets, futures exchanges, and major DeFi protocols in one codebase with a consistent handler pattern. The lazy-loading skills system is the right call for a project with this many optional dependencies — a missing SDK won't crash unrelated exchanges. The trade ledger with SHA-256 hashing and optional on-chain anchoring is a thoughtful addition; having an auditable trail of what the AI decided and why matters a lot when you're debugging a bad position after the fact.
This is a 12-day hackathon project that executes leveraged trades with real money, and the README doesn't include live performance data, drawdown figures, or documented behavior when the LLM returns something unexpected. The '118+ trading strategies' claim in one section and '4 strategies' in the summary table aren't reconciled anywhere. The 'swarm trading' feature — coordinating 20 wallets with Jito bundles on Pump.fun — is a market manipulation pattern presented without comment. Worth studying for its architecture and integration patterns; worth serious caution before running it live.
View on GitHub → Our full take →
That's the week. If you want these picks in your inbox before they hit the site, the email signup is at the bottom of the page — no noise, just the digest.
Get this in your inbox →