GitHub Trending Weekly 2026-02-18: Official AI Toolchains, Skills Ecosystem Forming, Backend Engineering Strikes Back
Data Period: 2026-02-11 ～ 2026-02-18 (Rolling 7 days) Sources: GitHub Trending weekly + monthly, GitHub Search API, HN Algolia
TL;DR: The biggest surprise this week is a Go language SQL tool—mickamy/sql-tap, which scored 231 points and 44 comments on HN, breaking through the wave of AI agent tools with pure engineering value. The weekly growth champion is Google's own langextract (+6,121 stars), while github/gh-aw's 302 HN points signal that GitHub itself is pushing AI agents into CI/CD. pi-mono, tambo, and gogcli simultaneously appear in weekly and monthly trends, showing clear signals of sustained popularity.
📈 Fastest Growing — Top 10 Star Growth This Week
Source:
github.com/trending?since=weekly🔁 = Also in monthly trends (Sustained popularity signal)
| # | Project | +Stars/Week | Total Stars | Language | Created |
|---|---|---|---|---|---|
| 1 | google/langextract | +6,121 | ★32,957 | Python | 2025-07 |
| 2 🔁 | badlogic/pi-mono | +3,326 | ★13,327 | TypeScript | 2025-08 |
| 3 🔁 | tambo-ai/tambo | +2,540 | ★10,641 | TypeScript | 2024-06 |
| 4 | Jeffallan/claude-skills | +2,461 | ★3,077 | Python | 2025-10 |
| 5 | danielmiessler/Personal_AI_Infrastructure | +2,263 | ★8,730 | TypeScript | 2025-09 |
| 6 🔁 | steipete/gogcli | +2,144 | ★4,008 | Go | 2025-12 |
| 7 | ChromeDevTools/chrome-devtools-mcp | +2,059 | ★25,839 | TypeScript | 2025-09 |
| 8 | github/gh-aw | +1,872 | ★3,107 | Go | 2025-08 |
| 9 | p-e-w/heretic | +1,778 | ★7,646 | Python | 2025-09 |
| 10 | steipete/summarize | +1,628 | ★3,598 | TypeScript | 2025-12 |
🆕 Top New Repos — Top 10 New Projects This Week
Source: GitHub Search API (
created:2026-02-11..2026-02-18, sorted by total stars)
| # | Project | Total Stars | Language | Created Date |
|---|---|---|---|---|
| 1 | zeroclaw-labs/zeroclaw | ★11,846 | Rust | 2026-02-13 |
| 2 | bwya77/vscode-dark-islands | ★3,571 | — | 2026-02-14 |
| 3 | HKUDS/ClawWork | ★1,921 | Python | 2026-02-15 |
| 4 | xyzeva/k-id-age-verifier | ★1,609 | TypeScript | 2026-02-11 |
| 5 | millionco/react-doctor | ★1,325 | TypeScript | 2026-02-13 |
| 6 | RevylAI/greenlight | ★1,060 | Go | 2026-02-11 |
| 7 | vercel-labs/portless | ★986 | TypeScript | 2026-02-15 |
| 8 | mickamy/sql-tap | ★888 | Go | 2026-02-14 |
| 9 | HKUDS/FastCode | ★820 | Python | 2026-02-13 |
| 10 | vigorX777/ai-daily-digest | ★756 | TypeScript | 2026-02-14 |
Weekly Focus — Fastest Growing Top 10
📈 #1 — google/langextract｜Google Steps In to Solve LLM Structured Extraction
A Python library for extracting structured information from unstructured text using LLMs with precise source grounding and interactive visualization.
Weekly +6,121 ★｜Total ★32,957｜Python｜Apache-2.0
LangExtract targets the most painful part of RAG pipelines: accurately extracting fields from unstructured text. The core selling point is "source grounding"—every extracted field can be traced back to its specific location in the original text, combined with an interactive visualization interface that makes manual verification intuitive. Supports Gemini API (gemini-flash, gemini-pro, etc.), has a PyPI package, and installs directly via pip install langextract.
Why it matters: Google building its own library for structured extraction signifies that this need is fundamental and universal enough that even the model provider finds it worth creating tooling for.
📈 #2 🔁 — badlogic/pi-mono｜Minimalist Coding Agent, Four Core Tools, 1k Token System Prompt
A TypeScript monorepo AI agent toolkit — coding agent CLI, unified LLM API, TUI library, vLLM pod manager.
Weekly +3,326 ★｜Total ★13,327｜TypeScript｜MIT｜🔁 Monthly Sustained Hit
pi-mono is a TypeScript monorepo by libGDX author Mario Zechner. Its core is a terminal coding agent CLI pi: just four core tools, a system prompt under 1,000 tokens, deliberately kept minimalist. The entire monorepo also includes a unified LLM API layer (pi-ai), a TUI library (pi-tui), Web UI components (pi-web-ui), and a vLLM pod manager (pi-pods).
Discussion on HN focuses on design philosophy: the author believes "what you leave out is more important than what you add," as well as pi's non-flickering TUI and turn rollback features. pi is also the underlying foundation for another AI coding tool in this week's Top New Repos.
📈 #3 🔁 — tambo-ai/tambo｜Generative UI SDK, Double Hit on Weekly + Monthly Charts (HN 101 points)
Generative UI SDK for React
Weekly +2,540 ★｜Total ★10,641｜TypeScript｜MIT｜🔁 Monthly Sustained Hit
Tambo allows AI agents to render corresponding React components directly based on conversation context, rather than just plain text responses. Version 1.0 scored 101 points on HN, with engineers discussing the core question: "When should we use Generative UI, and when should we still stick to fixed components?" Trending for two consecutive weeks, the signal is stronger than a single-week spike.
📈 #4 — Jeffallan/claude-skills｜66 Claude Code Skill Packs, AI Tool Plugin Market Takes Off
66 Specialized Skills for Full-Stack Developers. Transform Claude Code into your expert pair programmer.
Weekly +2,461 ★｜Total ★3,077｜Python｜MIT
claude-skills upgrades Claude Code from a general assistant to a "domain expert": security auditor, performance engineer, API designer... Each skill contains specialized system prompts and workflows. This points in the same direction as anthropics/skills and openai/skills appearing in monthly trends: Agent skills are forming as an independent ecosystem.
📈 #5 — danielmiessler/Personal_AI_Infrastructure｜Persistent AI Assistant Infrastructure That Knows Your Habits
An open-source personalized AI platform that knows your goals, history, and preferences across every session.
Weekly +2,263 ★｜Total ★8,730｜TypeScript｜MIT
Personal_AI_Infrastructure is the new work from Fabric author Daniel Miessler. Unlike stateless chatbots, it uses the "TELOS system" to record user goals, habits, and history in 10 Markdown files (MISSION.md, GOALS.md, PROJECTS.md, etc.), with a three-layer memory architecture (hot/warm/cold) allowing every conversation to pick up where the last left off. Natively based on Claude Code's hook system, supports ElevenLabs voice output and Discord notifications.
📈 #6 🔁 — steipete/gogcli｜Single CLI for Full Google Workspace Suite, Updated to v0.11
A fast, script-friendly command-line interface for the full Google Workspace suite.
Weekly +2,144 ★｜Total ★4,008｜Go｜🔁 Monthly Sustained Hit
gogcli covers all Google Workspace services with a single binary named gog: Gmail, Calendar, Drive, Docs, Slides, Sheets, Forms, Apps Script, Contacts, Tasks, Chat, Classroom, Keep. Using JSON-first output, it fits scripts and AI agents; supports multi-account management, OS keyring secure storage, and command whitelisting for AI agent sandboxing. v0.11.0 (2026-02-15) added Apps Script and Forms command groups. Installable via brew install steipete/tap/gogcli.
📈 #7 — ChromeDevTools/chrome-devtools-mcp｜Official Chrome MCP Server, Letting AI Agents Directly Control Browsers
Official Chrome DevTools MCP server for AI coding agents to control and inspect a live Chrome browser.
Weekly +2,059 ★｜Total ★25,839｜TypeScript｜Apache-2.0
chrome-devtools-mcp is the official MCP server from the Google Chrome DevTools team, allowing AI coding agents like Claude, Gemini, Cursor, and Copilot to directly control and inspect the browser via the Chrome DevTools Protocol. 26 tools cover input automation, navigation, performance analysis, network monitoring, and console debugging; underpinned by Puppeteer, launchable via npx chrome-devtools-mcp.
There have been multiple HN discussions, with criticism focused on "essentially a Puppeteer wrapper where the agent only sees the accessibility tree"; Addy Osmani defended it in a blog post as "eyes" for AI agents.
📈 #8 — github/gh-aw｜GitHub Officially Pushes "Continuous AI" into CI/CD (HN 302 points)
GitHub Agentic Workflows — actions, cai, ci, claude-code, codex, copilot
Weekly +1,872 ★｜Total ★3,107｜Go｜MIT
gh-aw is an official GitHub gh CLI extension that lets AI agents (one of GitHub Copilot, Claude Code, or OpenAI Codex) handle repo tasks directly in Actions workflows: code review, PR generation, issue triage, test fixing. GitHub calls this concept "continuous AI (cAI)".
HN 302 points, 142 comments—concentrated on two issues: the authorization boundary of AI agents in CI/CD (who approves the AI merge?), and whether this will make junior dev PR review work disappear.
📈 #9 — p-e-w/heretic｜Automatic Removal of Model Safety Alignment Using Directional Ablation (HN Buzz)
Automatic, fully parametrized censorship removal for transformer-based language models without retraining.
Weekly +1,778 ★｜Total ★7,646｜Python｜AGPL-3.0
heretic uses "directional ablation" combined with Optuna TPE parameter optimizer to automatically adjust direction vectors in each layer without retraining, minimizing refusal responses while minimizing KL divergence from the original model. Supports dense models, multimodal models, and multiple MoE architectures, supports bitsandbytes quantization to reduce VRAM requirements. There are already over 1,000 community derivative models based on heretic on Hugging Face.
HN Discussion (Nov 2025): The technical route is considered quite ingenious; discussion is split into two camps—where to draw the line between "removing political censorship" and "removing safety guardrails"?
📈 #10 — steipete/summarize｜Universal Content Summarizer for CLI + Chrome Side Panel
Summarize any URL, YouTube video, podcast, PDF, audio/video file, or RSS feed from CLI or browser sidebar.
Weekly +1,628 ★｜Total ★3,598｜TypeScript
summarize is another open source tool from gogcli author Peter Steinberger: CLI supports URL, YouTube, Podcast, PDF, audio/video, RSS; Chrome Side Panel version (v0.11+) adds streaming conversation agents and history recording. Video summarization includes slide extraction—OCR + timestamped screenshot cards. Prioritizes published captions, with Whisper as fallback. Supports OpenAI-compatible local endpoints and OpenRouter. Installable via brew install (macOS arm64).
Weekly Focus — Top New Repos Top 10
🆕 New #1 — zeroclaw-labs/zeroclaw｜5 Days ★11,846, New High for Rust AI Assistant Framework
Fast, small, and fully autonomous AI assistant infrastructure — deploy anywhere, swap anything 🦀
★11,846｜Rust｜Created 2026-02-13
zeroclaw amassed nearly 12k stars within 5 days of creation, with 1,189 forks. Design philosophy is "Zero Overhead + Fully Autonomous + Swappable Components"—models, memory backends, and tool layers are all hot-pluggable, deployable on cloud or edge devices. It is the Rust-based AI assistant framework with the highest star count at birth so far.
🆕 New #2 — bwya77/vscode-dark-islands｜Bringing JetBrains Islands Dark Visual Style to VS Code
A dark VS Code color theme replicating JetBrains' Islands Dark: floating glass panels, rounded corners, smooth animations.
★3,571｜PowerShell + Shell｜MIT｜Created 2026-02-14
vscode-dark-islands replicates the visual language of the Islands Dark theme launched by JetBrains in September 2025, using a Custom UI Style extension to inject CSS for floating glass panels and rounded corner effects that exceed standard theme limits. Includes one-click install scripts (Unix/macOS and Windows), defaulting to IBM Plex Mono (editor) + FiraCode Nerd Font (terminal). Zero to 3,500+ stars in a week, driven mainly by developer community social media spread.
🆕 New #3 — HKUDS/ClawWork｜Evaluating AI Agent "Real Workplace Productivity" with $10 Simulation (HN Buzz)
Economic benchmark: give an AI agent $10 and 44 occupational tasks — measure real income earned per token spent.
★1,921｜Python｜MIT｜Created 2026-02-15
ClawWork is an AI agent economic evaluation framework from the HKUDS lab at HKU (same team as LightRAG). Agents are given $10 simulated funds and 220 real occupational tasks (GDPVal dataset, covering 44 economic sectors), earning income by task quality and paying by token consumption—forcing agents to make strategic trade-offs between "doing tasks now" and "investing in learning first". Comes with a React + WebSocket real-time economic dashboard, supporting multi-model arenas like GPT-4o and Claude.
HN Discussion: The evaluation framework design is considered closer to actual deployment scenarios than traditional benchmarks, though some question the reliability of LLM quality assessment.
🆕 New #4 — xyzeva/k-id-age-verifier｜Discord, Twitch, Snapchat Age Verification Bypass Tool (HN Discussion)
Automates age verification on platforms using k-id by replicating its AES-GCM encrypted facial metadata payload.
★1,609｜TypeScript｜Created 2026-02-11
k-id-age-verifier generates legitimate-looking facial metadata payloads by replicating the AES-GCM encryption protocol of k-id (facial recognition age verification service used by Discord, Twitch, Kick, Quora, Snapchat), without needing actual face scans. Currently in an "attack and defense" loop with k-id—k-id has patched multiple times, and maintainers continue to update bypass methods.
HN 302 points discussion and 404 Media reporting focus on a core issue: When age verification is tied to biometrics (face scans), who protects user privacy data? The technology itself is neutral, but it has sparked broad discussion on the legality of platforms mandating biometric collection.
🆕 New #5 — millionco/react-doctor｜Let AI Agents Be Your React Doctor
Let coding agents diagnose and fix your React code
★1,325｜TypeScript｜MIT｜Created 2026-02-13
The Million.js team (authors of the famous React performance optimization package) launched react-doctor, allowing AI agents to automatically diagnose React code issues: component performance, incorrect hook usage, accessibility, etc. Topics include skill, designed as an agent skill to be directly integrated into workflows like Claude Code.
🆕 New #6 — RevylAI/greenlight｜Compliance Scanner Before iOS App Store Submission
Pre-submission compliance scanner for iOS apps: detect common App Store rejection reasons before you submit.
★1,060｜Go｜MIT｜Created 2026-02-11
greenlight lets iOS developers run greenlight preflight before App Store submission, running four scanners in parallel: 30+ code pattern detections (private API calls, hardcoded secrets, payment bypass, missing ATT prompt), privacy manifest verification (PrivacyInfo.xcprivacy), compiled IPA file analysis, and App Store Connect API remote metadata confirmation. Outputs JSON and JUnit format, CI/CD friendly; supports Claude Code and Codex skill integration for automatic issue fixing.
🆕 New #7 — vercel-labs/portless｜.localhost Named URLs Designed for Humans and Agents
Replace port numbers with stable, named .localhost URLs. For humans and agents.
★986｜TypeScript｜Apache-2.0｜Created 2026-02-15
portless replaces port number URLs like localhost:3000 with stable myapp.localhost, so AI agents calling local services don't need to memorize port numbers or break chains due to restarts. From Vercel Labs, the description specifically emphasizes "For humans and agents"—agent reachability in local development environments is starting to be taken seriously by major players.
🆕 New #8 — mickamy/sql-tap｜Biggest Surprise of the Week: Pure Engineering Tool Hits HN 231 Points Amidst AI Wave
Watch SQL traffic in real-time with a TUI
★888｜Go｜MIT｜Created 2026-02-14
sql-tap is a terminal TUI tool written in Go that intercepts and displays PostgreSQL and MySQL SQL query traffic in real-time without modifying application code. Scored 231 points and 44 comments on HN on its creation day.
This is particularly conspicuous in a week trending with AI agent tools. Discussion focused on: lighter than pgAdmin's query analysis, more focused than Wireshark, especially useful for debugging N+1 problems and slow queries. Pure engineering problem, zero AI packaging, and it exploded just like that.
🆕 New #9 — HKUDS/FastCode｜Claims Codebase Understanding Framework 3-4x Faster, 44-55% Cheaper than Cursor, Claude Code
Token-efficient framework for code understanding in large codebases: hierarchical indexing + semantic search + relationship graphs.
★820｜Python｜MIT｜Created 2026-02-13
FastCode also comes from the HKUDS lab, targeting Q&A and navigation tasks for large codebases. Three-stage architecture: hierarchical code indexing (file → class → function → documentation), semantic structured representation (embedding + BM25), relationship graph modeling (call graph, dependency graph, inheritance graph). Supports AST parsing for 8+ languages, offers Web UI, REST API, and CLI interfaces. The paper claims to outperform Cursor and Claude Code on benchmarks like SWE-QA, but independent verification is pending community follow-up.
🆕 New #10 — vigorX777/ai-daily-digest｜Zero-Dependency Bun Script, AI Curates 90 Top Tech Blogs Daily
Zero-dependency TypeScript/Bun script that scrapes 90 curated tech blogs, AI-scores articles, and generates a structured daily Markdown digest with trend analysis.
★756｜TypeScript｜Created 2026-02-14
ai-daily-digest scrapes 90 top tech blogs from Andrej Karpathy's curated list (10-way concurrency, 15s timeout), uses AI to score and filter from three dimensions, and generates a daily Markdown digest containing Mermaid pie charts, ASCII bar charts, and translated Chinese titles. Six article categories: AI/ML, Security, Engineering, Tools/Open Source, Opinions, Other. AI backend supports Gemini API and any OpenAI-compatible endpoint (including DeepSeek). Pure TypeScript single file, no third-party dependencies, runs on Bun native fetch.
Monthly Trend Comparison
Three projects in this week's weekly list also appearing in monthly trends (🔁):
| Project | +Stars This Week | Monthly Rank Direction | Sustained Theme |
|---|---|---|---|
| badlogic/pi-mono | +3,326 | Monthly Sustained Hit | AI agent toolkit full stack |
| tambo-ai/tambo | +2,540 | Monthly Sustained Hit | Generative UI |
| steipete/gogcli | +2,144 | Monthly Sustained Hit | Google Suite CLI |
Other notable signals in monthly trends: anthropics/skills and openai/skills are both in monthly trends, echoing this week's claude-skills breakout—"AI Skills Market" is one of the strongest recurring themes this month.
Weekly Trend Insights
1. Officialization of AI Toolchains: The three most important signals this week come from big companies stepping in—Google (LangExtract), Google Chrome Team (chrome-devtools-mcp), GitHub Official (gh-aw). AI tools are no longer just a community playground; platforms are building "official channels".
2. Skills as the New App Store for AI Agents: claude-skills (+2,461), anthropics/skills, and openai/skills all in monthly trends, plus nicobailon/visual-explainer and MooseGoose0701/skill-compose—"Pluggable Skill Packs" are becoming the distribution format for this generation of AI agent applications.
3. The Counterattack of Backend Tools: sql-tap (HN 231 points) reminds us: real engineering problems always have a market. In a weekly trend dominated by AI agent tools, a pure SQL monitoring tool breaking out on engineering quality and clear problem definition is the most thought-provoking contrast this week.
