Pulse Weekly #2, Anthropic Week, Git Wins

2026-04-14 | 10 min read


What I found interesting from Hacker News and GitHub Trending for the week of April 7-14, 2026. For HN, I only looked at stories with 100+ points. For GitHub, I checked trending pages across 8 language categories (All, Go, Python, Rust, TypeScript, JavaScript, Java, C++) for daily, weekly, and monthly, then removed duplicates and compared.

Note: HN scores and GitHub star counts were captured on April 14, 2026. Both fluctuate over time.

#The Big Picture

HN had a big week for Anthropic, Git workflows, and developer tools. Git commands was the #1 story. On GitHub, the Claude Code plugin ecosystem blew up (again). Both platforms had strong AI and developer tool content, but from different angles.

#1. Anthropic Is the Main Character

Both platforms agree: Anthropic is where developer attention is this week. But for very different reasons.

On HN, Anthropic took over the AI discussion with both wins and backlash:

PtsCommentsTitle
1541835Project Glasswing: Securing critical software for the AI era
1271340Small models also found the vulnerabilities that Mythos found
845659System Card: Claude Mythos Preview (pdf)
746650Pro Max 5x quota exhausted in 1.5 hours despite moderate usage
612602ML promises to be profoundly weird
581140Exploiting the most prominent AI agent benchmarks
546421Anthropic downgraded cache TTL on March 6th
518427AI assistance when contributing to the Linux kernel
465359Claude mixes up who said what
432200I’ve been waiting over a month for Anthropic to respond to my billing issue

Glasswing (1541 pts) was the week’s #3 story. Four other Anthropic-related posts also made the front page: quota limits, cache TTL changes, attribution behavior, and billing support. Berkeley researchers showed AI benchmarks can be tricked (581 pts), and smaller models found the same bugs as Mythos (1271 pts).

On GitHub, the Claude Code ecosystem blew up. The top repos are almost all Claude Code plugins, skills, and wrappers:

RepoStarsMonthlyWeeklyDaily
NousResearch/hermes-agent, self-improving agent84K+69,018+48,286+8,282
thedotmack/claude-mem, memory plugin for Claude Code56K+20,099+6,570+2,979
farion1231/cc-switch, multi-agent CLI assistant45K+16,551+4,439+649
forrestchang/andrej-karpathy-skills, single CLAUDE.md file33K+17,646+14,317+9,230
shanraisshan/claude-code-best-practice, from vibe coding to agentic engineering44Kn/an/a+2,569
gsd-build/get-shit-done, meta-prompting system53K+22,926n/a+1,015
multica-ai/multica, managed agents platform12Kn/a+8,580+1,640
coleam00/Archon, harness builder for AI coding18K+3,840+3,647+452
TheCraigHewitt/seomachine, SEO Machine for Claude Code6,189n/a+2,548n/a

The cross-period numbers tell a story: hermes-agent is speeding up (48K of its 69K monthly came this week). andrej-karpathy-skills just had a spike (+9.2K in one day). cc-switch might be cooling off (daily is slowing down vs weekly).

The interesting part: HN had active discussions about Claude Code quota limits (746 pts) and cache TTL changes (546 pts) the same week GitHub had the biggest wave of Claude Code tooling ever. Lots of discussion and lots of building at the same time. A single CLAUDE.md text file (andrej-karpathy-skills) got 14K stars in a week.

#2. AI Agents and MCP Are Spreading

New projects keep using the agent pattern. MCP integrations are showing up everywhere on GitHub, while HN argues about whether the standard should even exist.

On HN, the MCP vs Skills debate got 458 pts and 373 comments:

PtsCommentsTitle
458373I still prefer MCP over skills
446323OpenAI backs Illinois bill that would limit when AI labs can be held liable
617263GLM-5.1: Towards Long-Horizon Tasks
392367Muse Spark: Scaling towards personal superintelligence

On GitHub, MCP is winning through implementation:

RepoStarsMonthlyWeeklyDaily
aaif-goose/goose, open-source AI developer agent42K+9,050+4,046n/a
HKUDS/DeepTutor, ReasoningAgent for papers18K+7,082+6,210n/a
googleapis/mcp-toolbox, Google’s MCP Toolbox15K+1,191+626n/a
LaurieWired/GhidraMCP, MCP for Ghidra reverse engineering8,382n/a+211n/a
tailcallhq/forgecode, open-source AI coding agent6,534+1,722+602n/a
ChromeDevTools/chrome-devtools-mcp, DevTools for coding agents35K+6,162n/a+243
zinja-coder/jadx-ai-mcp, MCP server for JADX1,765n/a+110+11

Goose is growing steady, not spiking. DeepTutor just took off recently. Four of seven repos here are MCP integrations (Google, Chrome DevTools, Ghidra, JADX), HN argues about the standard, GitHub builds integrations.

#3. Git and Developer Tools Still Win

The #1 story of the entire week on HN was about git commands. Not AI, not a product launch.

On HN:

PtsCommentsTitle
2327507Git commands I run before reading any code
862491GitHub Stacked PRs
802455A new spam policy for “back button hijacking”
468148Servo is now available on crates.io
468143The peril of laziness lost
462396jj, the CLI for Jujutsu
425264Make tmux pretty and usable (2024)
41048USB for Software Developers: An introduction to writing userspace USB drivers
378119S3 Files
373166Lean proved this program correct; then I found a bug
407280The economics of software teams: Why most engineering orgs are flying blind
361304We have a 99% email reputation, but Gmail disagrees

What stands out: GitHub Stacked PRs (862 pts) is something developers have wanted for years. And jj/Jujutsu (462 pts) keeps coming back, the Git alternative has real staying power.

On GitHub, git tools keep growing:

RepoStarsWeeklyDaily
jesseduffield/lazygit, terminal UI for git58K+711n/a
gitbutlerapp/gitbutler, Git branch management (Rust)20K+356n/a

lazygit has been around since 2018 and is still growing. Combined with the HN #1 story about git commands, git workflow tools just never stop being useful.

#4. Rust Everywhere

Rust is showing up across storage, AI tools, dev tools, and consumer apps. Not just systems programming anymore.

On HN, Servo’s first crate on crates.io made the front page:

PtsCommentsTitle
468148Servo is now available on crates.io

On GitHub, seven Rust repos are trending across different categories:

RepoStarsMonthlyWeeklyDaily
rustfs/rustfs, S3-compatible object storage26K+2,635+1,317+212
mayocream/koharu, ML-powered manga translator3,221n/a+1,078+221
dmtrKovalenko/fff.nvim, file explorer for Neovim4,817+3,616+867n/a
gitbutlerapp/gitbutler, Git branch management20Kn/a+356n/a
tonhowtf/omniget, desktop video downloader1,643n/a+472+133
pola-rs/polars, Rust-powered dataframes38Kn/a+158n/a
biomejs/biome, web dev toolchain24Kn/a+147n/a

fff.nvim peaked earlier and is slowing down. koharu (manga translator) is still picking up speed. rustfs has steady numbers across all periods, looks like real adoption, not a viral spike.

#5. AI Finance Is Having a Moment

This is a GitHub-only trend this week, HN had almost nothing about AI trading.

RepoStarsMonthlyWeeklyDaily
virattt/ai-hedge-fund, AI hedge fund team54Kn/a+2,670+1,007
TauricResearch/TradingAgents, multi-agent trading framework50K+18,422n/an/a
shiyu-coder/Kronos, foundation model for financial markets18K+5,766n/a+965
HKUDS/AI-Trader, AI trading agent13Kn/a+1,129n/a
tradingview/lightweight-charts, financial chart library15Kn/a+895n/a

TradingAgents peaked earlier and is cooling down (only shows in monthly). Kronos just started getting attention (high daily, low monthly). ai-hedge-fund looks like steady, ongoing interest.

#6. LLM Infrastructure

The tools that process, serve, and run models keep growing quietly. No spikes, just steady growth. This is a GitHub-only trend this week, HN had no front page stories about LLM infra.

RepoStarsMonthlyWeeklyDaily
microsoft/markitdown, convert files to Markdown for LLMs108K+16,296+13,336+1,672
ollama/ollama, run LLMs locally169Kn/a+1,490n/a
ggml-org/llama.cpp, LLM inference in C/C++104K+6,240+1,758+172
google-ai-edge/LiteRT-LM, on-device LLM runtime3,748+2,720+1,779n/a
mem0ai/mem0, memory layer for AI agents53Kn/a+903n/a

markitdown had almost all its monthly growth this week (82%). LiteRT-LM is speeding up, Google’s on-device runtime is getting more attention. llama.cpp had a spike earlier in the week (daily is lower than the weekly average).

#7. Show HN Highlights

Developer projects that shipped this week:

PtsCommentsTitle
511295Show HN: boringBar, a taskbar-style dock replacement for macOS
252207Show HN: Stop paying for Dropbox/Google Drive, use your own S3 bucket instead
240107Show HN: Oberon System 3 runs natively on Raspberry Pi 3 (with ready SD card)
23428Show HN: Gemma 4 Multimodal Fine-Tuner for Apple Silicon
185127Show HN: I built a social media management tool in 3 weeks with Claude and Codex
177169Show HN: I built a Cargo-like build tool for C/C++
15538Show HN: FluidCAD, Parametric CAD with JavaScript
14034Show HN: Marimo pair, Reactive Python notebooks as environments for agents
11844Show HN: Claudraband, Claude Code for the Power User

boringBar (511 pts) got a lot of interest, macOS dock replacements are clearly popular. Locker.dev (252 pts) lets you use your own S3 bucket instead of paying for Dropbox, another open-source alternative getting attention. A Cargo-like build tool for C/C++ (177 pts, 169 comments) got a good discussion going about what C/C++ build tooling should look like.

#Quick Hits

From HN:

From GitHub:

#Week-Over-Week: PW1 (Apr 7) vs PW2 (Apr 14)

Comparing GitHub Trending data between the two weeks.

#My Take

The #1 story on HN (2327 pts) was about git commands. Not AI, not a product launch, just a developer sharing how they read code. Developer tools, tmux guides, and USB driver tutorials keep beating product announcements. HN still likes real stuff over hype.

Anthropic was everywhere this week. Glasswing and Mythos got a lot of attention, and so did discussions about quota limits, cache TTL changes, and billing support. On GitHub, a whole new wave of Claude Code tools showed up: memory plugins, workflow wrappers, prompt templates, skills files. HN discusses how the tools work, GitHub builds on top of them.

The star numbers are huge. hermes-agent got 48K stars in a week. andrej-karpathy-skills got 14K. GitHub Trending right now is a mix of fast-growing Claude Code plugins and steady growth from tools that have been around for a while (llama.cpp, ollama, Godot, Rust in general).

What actually stays the same week to week: Google’s on-device AI stack keeps growing quietly, Goose has the most stable growth in the AI agent space, and git workflow tools just never stop being useful.


If this was useful, consider buying me a coffee :)