Forkcast Hacker News #1: ymawky, let-go, rust-but-lisp, ml-sharp-web, trust, ableton-live-mcp

Episode Summary#

The first Hacker News roundup explores six eclectic projects from the HN community. From a web server written entirely in ARM64 assembly to a Clojure-like language in Go with 50x faster startup than JVM Clojure, an experimental Rust-Lisp fusion language, and an MCP server that lets AI agents control Ableton Live for music production.

Repositories Covered#

ymawky#

A static file web server for macOS written entirely in ARM64 assembly. Supports GET, PUT, DELETE, HEAD, OPTIONS, Range headers, percent-encoded URLs, and directory listing with mitigations against slowloris attacks. A remarkable demonstration of low-level systems programming.

let-go#

An open-source programming language written in pure Go with ~90% syntax compatibility with JVM Clojure. Compiles to a ~10MB static binary with 7ms cold start — approximately 50x faster than JVM Clojure — making it practical for CLI tools and microservices.

rust-but-lisp#

An experimental programming language project that blends Rust’s strong typing and memory safety with Lisp-style macros and functional programming expressiveness. Explores the design space between systems programming and metaprogramming.

ml-sharp-web#

A web framework for building machine learning applications in F#/.NET. Brings ML model serving, data processing, and web API development together in a functional-first .NET environment.

trust#

A distributed trust and reputation system for peer-to-peer networks. Implements decentralized identity verification and confidence scoring to enable secure interactions without centralized authorities.

ableton-live-mcp#

An MCP (Model Context Protocol) server that gives AI agents the ability to control Ableton Live — creating clips, adjusting parameters, arranging tracks, and generating music through programmatic control of the digital audio workstation.

re_gent#

AI agents generate hundreds of file changes across dozens of turns with no audit trail. When an agent breaks something, you can’t trace which prompt or tool call caused it. Agent VCS with BLAKE3 content-addressed snapshots. DAG-based step history in .regent/ directory. SQLite index with sub-10ms lookups. Per-session branches with concurrent refs. Blame shows which step, session, tool, and prompt touched each line. Hook-driven integration with Claude Code, Codex, OpenCode. VSCode extension with inline blame. Apache 2.0.

agent-desktop#

Native desktop automation CLI for AI agents

php-fts#

Adding search to a PHP app means Elasticsearch, Meilisearch, or MySQL FULLTEXT — all requiring extra services, daemons, or extensions. Shared hosting gives you none of that. Trigram indexing with O(1) lookup via ~810 KB fixed-size index (37^3 entries). BM25 + IDF scoring (k1=1.5, b=0.75). Median search: 3.2ms, P95: 12.5ms on 10K docs. Bulk insert 2.4x faster. Field boosting, highlighting, soft deletes, compaction. Runs on PHP 8.1+ with zero extensions. 50K docs = 106 MB index. MIT license.

nfsdiag#

Debugging NFS issues means running showmount, rpcinfo, mount, and reading /proc by hand. There’s no single tool that tests every NFS subsystem from the client side and produces a structured report. Modular C architecture. Tests rpcbind port 111, NFS port 2049, NFS v2/v3/v4/v4.1/v4.2 via RPC NULLPROC. Auto-mounts with v4.2->v4.1->v4->v3 fallback. POSIX ACLs, NFSv4 ACLs, SELinux contexts. copy_file_range, fallocate, O_DIRECT I/O tests. fcntl locking, root_squash detection, pNFS+RDMA checks. JSON and standalone HTML output. 14 Docker failure fixtures. No license specified.

agent-skills-eval#

You write a skill for your AI agent but have no idea if it actually improves outputs. Everyone ships skills with zero empirical evidence they work. Runs skills against prompts twice — with_skill and without_skill — judge model grades both side by side. Pass/fail with cited assertions. Tool-call assertions for deterministic checks. OpenAI-compatible provider support. Portable JSON+JSONL artifacts, static HTML reports. Implements full agentskills.io spec. Concurrency configurable (default 4). npm: npx agent-skills-eval. MIT license.

rvcore#

RISC-V emulator that runs DOOM

stage-cli#

Reviewing AI-generated code changes means scrolling through a massive unified diff with no structure. You can’t tell which changes go together or what to focus on first. Organizes local changes into reviewable chapters, opens browser UI. Works with any AI agent. npm: stagereview. /stage-chapters slash command in Claude Code. Diff scopes: work (staged+unstaged+untracked), staged, unstaged. pnpm workspaces, Vitest, Biome. Everything runs on your machine. 72 commits. MIT license.

adamsreview#

Claude Code’s built-in review misses real bugs and flags false positives. You need ensemble reviews with multiple models and automated fix loops. 6 slash commands: review (7 parallel sub-agent lenses), codex-review (peer), add (inject external findings), walkthrough (interactive AskUserQuestion driver), fix (auto-fix loop with re-review), promote (human override). Ensemble mode adds Codex CLI pass + PR bot scrape. Finding lanes: deep-manual, deep-report, light. State in ~/.adams-reviews/. 228 commits. MIT license.

engine#

Durable workflow engine in Rust, one binary, Postgres or SQLite

kstack#

Skill pack for Claude Code that helps monitor and troubleshoot K8s clusters

pii-shield#

Zero-code K8s sidecar for log sanitization. Detects secrets via Entropy Analysis

undoredo#

Undo-redo in Rust using deltas, snapshots or commands

light-crime-audio-player#

The greatest winamp-inspired native macOS audio player ever

kanban-cli#

One line ui for markdown kanban files

liteflow#

1000-line C program where an LLM can edit the DAG mid-run using a four-verb grammar

serenedb#

The First Real-Time Search Analytics Database

modafinil#

Let your agents run while your MacBook lid is closed

better-design#

Open-source design systems for shadcn/ui. 27 themed component libraries

chuchu#

Android SSH clients have terrible terminal emulation, no image support, and clunky keyboard handling. You can’t get a proper kitty-compatible terminal on your phone. Kotlin + Jetpack Compose UI, Zig JNI bridge, Ghostty VT terminal engine, libssh2 + OpenSSL for SSH. Kitty image protocol support. 400+ themes from Ghostty repo. Tailscale + SSH password/key auth. Configurable accessory keys. v0.2.1, 10 releases, 106 commits. MIT license.

platos#

Runtime infrastructure to deploy, monitor and manage agents in production

perkins#

ASCII pixel art editor for the terminal

concord#

A feature-rich TUI client for Discord, written in Rust

ant#

JavaScript for ants, a tiny runtime with big ambitions

DiffCAD#

Review FreeCAD model changes like code

agent-sanbox#

Runs AI coding agents inside isolated Docker containers

mac-juice-monitor#

Keep track of all your Bluetooth-connected devices’ battery levels in the macOS menu bar

Watch#

• Video: https://www.youtube.com/watch?v=vRjzTzlW2rY
• Cover: https://i.ytimg.com/vi/vRjzTzlW2rY/hqdefault.jpg

Notes#

Transcript and notes will be added from Forkcast output artifacts.