GitHub Hot — 5 April 2026

Top 10 repos trending on GitHub this week — what they do, why they matter, and how to use them in your projects.

1. ultraworkers/claw-code

170,681 stars this week · Rust

Claw Code is a Rust-based CLI agent harness that appears to be a Claude Code clone/alternative, and it went viral purely on meme momentum after becoming the fastest repo to 100K stars — treat it with extreme skepticism.

Use case

This repo is almost certainly a viral stunt or placeholder that gamed GitHub trending by leveraging the Claude Code hype cycle. The real problem it claims to solve — a Rust-native CLI harness for AI coding agents — is legitimate, but there is no evidence this implementation is production-ready or trustworthy. Do not build on this; the star count is the product, not the software.

Why it's trending

It hit 100K stars in record time by riding the Claude Code viral moment and Discord community coordination, not organic developer adoption. It's trending because GitHub's trending algorithm is gameable and the developer community is watching the spectacle, not because the code ships value.

How to use it

Do NOT install or run this in any environment with real credentials or production access — provenance is unverified.
If you're genuinely curious about the Rust agent harness concept, audit rust/ locally in a sandbox: git clone https://github.com/ultraworkers/claw-code && cd claw-code/rust && cargo build --release
Run ./target/release/claw doctor as the docs suggest to see what health checks actually pass.
Cross-reference PARITY.md to understand what percentage of claimed functionality is actually ported vs. aspirational.
Treat any auth or session flows as untrusted until the project has a real release history and security audit.

How I could use this

Write a blog post titled 'How 100K Stars Lied to Me' — do a live teardown of claw-code's commit history, star-growth rate, and actual code completeness vs. a legitimate project like gitoxide or Zed. This kind of critical analysis gets bookmarked by senior devs and is great for establishing credibility.
Build a small 'GitHub Hype Detector' career tool using the GitHub API + a Supabase table that tracks star velocity, contributor count, commit frequency, and issue close rate for any repo — surface a 'legitimacy score' so you can quickly sanity-check viral repos before adding them to your resume's tech stack.
Use the genuine underlying concept (a local CLI agent harness) as inspiration: wire up Vercel AI SDK's useChat to spawn a sandboxed agent on your blog that can answer questions about your own codebase by reading a pre-indexed vector store in Supabase pgvector — no sketchy third-party binary required.

2. Gitlawb/openclaude

16,326 stars this week · TypeScript

OpenClaude is a free, open-source drop-in replacement for Anthropic's Claude CLI that routes to any LLM backend — OpenAI, Gemini, Ollama, DeepSeek, etc. — using one unified terminal interface.

Use case

Developers locked out of Claude's paid CLI tier (or wanting to swap providers without relearning tooling) can use OpenClaude to keep the same agent workflow — file edits, bash execution, grep, MCP tools — while pointing at a free or cheaper backend like Ollama locally or GitHub Models for zero-cost API calls. Example: you're iterating on a Next.js component at 2am, your Claude credits are burned, you just swap the provider profile to Ollama's local Llama 3 and keep working without changing your muscle memory.

Why it's trending

It surged this week because Anthropic's official Claude Code CLI is paywalled and usage-limited, and developers are actively seeking open alternatives that don't require an Anthropic subscription. The timing aligns with broader frustration over AI API cost spikes across providers.

How to use it

Install globally: npm install -g @gitlawb/openclaude,2. Launch the CLI: openclaude — on first run it drops you into an interactive shell,3. Set up your preferred provider (e.g. Ollama for free local inference): type /provider inside the CLI and follow the guided prompts to save a named profile,4. For zero-cost cloud usage, run /onboard-github to configure GitHub Models (free tier available with a GitHub account, no credit card),5. Use agent commands directly: e.g. type refactor src/components/BlogCard.tsx to use server components and it will read, edit, and confirm file changes inline with streaming output

How I could use this

Wire OpenClaude into Henry's blog repo as a local dev assistant: create a shell alias blog-ai that launches openclaude scoped to the /posts directory, so he can prompt 'improve SEO metadata for this post' and it edits MDX frontmatter directly — no copy-pasting into a chat UI.
Build a career-tool CLI workflow: point OpenClaude at a /career folder containing his resume JSON and a job-description.txt, then prompt it to generate a tailored cover letter and diff his resume bullets against the JD keywords — all file-native, version-controllable, and swappable between GPT-4o and DeepSeek depending on cost.
Use OpenClaude's MCP (Model Context Protocol) support to connect it to his Supabase blog database: expose a simple MCP tool that queries post analytics, then prompt 'which of my posts has the highest read-time but lowest traffic — suggest a rewrite angle' to get data-grounded content strategy without leaving the terminal.

3. VoltAgent/awesome-design-md

14,587 stars this week · HTML · awesome-list design-md design-system design-tokens

A curated library of DESIGN.md files that codify real websites' design systems in plain markdown so AI coding agents can generate pixel-perfect, consistent UI without Figma exports or JSON schemas.

Use case

When you ask Cursor, Copilot, or Claude to build a new page, it invents arbitrary colors, spacing, and typography that clash with your existing UI. Dropping a DESIGN.md into your project root gives the agent a single source of truth — for example, placing a Vercel-inspired DESIGN.md lets you prompt 'build me a pricing page' and get back the exact neutral grays, Inter font stack, and border-radius tokens Vercel uses, not a random Bootstrap guess.

Why it's trending

Google Stitch just introduced the DESIGN.md spec this week, creating a new standard file convention analogous to AGENTS.md — this repo is the first large curated library of real-world examples, so it's riding the first-mover wave while the format is still being defined.

How to use it

Browse the repo and pick a DESIGN.md that matches your target aesthetic (e.g., Linear, Vercel, or Stripe).
Copy the file into your project root: curl -o DESIGN.md https://raw.githubusercontent.com/VoltAgent/awesome-design-md/main/designs/linear.md
Open your AI agent (Cursor, Windsurf, Claude Code) and prompt it with context: 'Read DESIGN.md and build a blog post card component that follows this design system exactly.'
For Supabase/Next.js projects, extend the DESIGN.md by appending your own tokens — e.g., add a ## Custom Tokens section with your Supabase green accent and Tailwind config overrides.
Commit DESIGN.md to your repo so every future agent session and every team member's AI tool shares the same design contract.

How I could use this

Create a custom DESIGN.md for Henry's blog by starting from the Linear or Vercel template, then overriding the color palette with his personal brand colors and adding a ## Blog-Specific Components section that defines how post cards, code blocks, and author bylines should look — then every new page he scaffolds with Cursor will be instantly on-brand without manual style corrections.
Build a 'Portfolio Mode' toggle for his blog: maintain two DESIGN.md variants (one editorial/blog-focused, one recruiter-facing/clean-minimal), and wire a Next.js middleware flag that swaps which one is active — then use an AI agent to regenerate the resume/cover-letter pages using the recruiter-facing design system on demand.
Add a DESIGN.md-aware AI feature to his blog's admin dashboard: parse his DESIGN.md at build time with a server action, feed its color/typography tokens into an OpenAI prompt, and auto-generate og:image social cards for each new post that are guaranteed to match his site's exact visual identity rather than using a generic template.

4. claude-code-best/claude-code

13,880 stars this week · TypeScript

A reverse-engineered, fully debuggable TypeScript reconstruction of Anthropic's Claude Code CLI — with type fixes, OpenAI-compatible endpoints, and enterprise monitoring hooks baked in.

Use case

Anthropic ships Claude Code as a minified, opaque binary with no source you can inspect or extend. This repo gives you the full readable TypeScript source so you can fork the CLI, swap in your own API endpoint (e.g., a self-hosted OpenAI-compatible proxy), add custom auth, or instrument it with your own Sentry project — without being locked into Anthropic's billing or feature flags. Concrete example: you want Claude Code to hit your own LLM proxy that load-balances between Claude and GPT-4o depending on task type.

Why it's trending

Claude Code just hit mainstream adoption as a terminal-native AI coding agent, but its closed binary frustrated developers who wanted to customize or self-host it. This repo landed V5 with OpenAI-compatible routing and custom login — making it the first viable 'bring your own backend' drop-in replacement, which is exactly what cost-conscious teams want right now.

How to use it

Install Bun if you don't have it: curl -fsSL https://bun.sh/install | bash,2. Clone and install deps (lock file is pinned, so this is deterministic): git clone https://github.com/claude-code-best/claude-code && cd claude-code && bun i,3. Copy the env template and point it at your API: cp .env.example .env then set ANTHROPIC_API_KEY or set OPENAI_API_BASE to your proxy URL if using OpenAI-compatible mode,4. Start in dev mode with full source maps and hot reload: bun run dev,5. To wire in your own auth flow, edit the custom login module documented at https://ccb.agent-aura.top/docs/features/custom-platform-login — this is where you'd inject a Supabase JWT instead of an Anthropic token

How I could use this

Build a 'blog post draft agent' by forking this CLI and adding a custom slash command /draft that reads your Supabase posts table schema, pulls your last 5 published posts as style examples, then generates a new draft and inserts it as a row — all from your terminal without touching the browser.
Wire the custom login endpoint to your Supabase Auth so that when Henry runs the CLI on any machine, it authenticates via his existing blog's user session — meaning one set of credentials gates both the blog CMS and the AI coding agent, no separate Anthropic account needed for teammates or contractors.
Use the OpenAI-compatible routing layer to A/B test models on different coding tasks in your blog's AI features: route 'explain this code' requests to Claude Haiku (cheap) and 'refactor this component' requests to GPT-4o, logging latency and token cost per task type to a Supabase table — giving you real data to decide which model to pay for.

5. openai/codex-plugin-cc

11,896 stars this week · JavaScript

An OpenAI-official Claude Code plugin that lets you run Codex code reviews and delegate background tasks without leaving your Claude Code workflow.

Use case

Developers using Claude Code as their primary AI coding environment previously had to context-switch to a separate Codex session to get a second AI opinion on their code. This plugin lets you run /codex:review on uncommitted changes or branch diffs directly inside Claude Code — for example, before merging a Supabase RLS policy change, you can get an adversarial review that actively tries to poke holes in your logic without opening a new tool.

Why it's trending

This dropped right as the OpenAI vs Anthropic agentic coding war is peaking — developers are actively picking sides between Codex and Claude Code, and this plugin is OpenAI's move to stay embedded in Claude Code users' workflows rather than lose them entirely. 11k+ stars in a week signals the 'use both AIs together' workflow hit a nerve.

How to use it

Inside Claude Code, run /plugin marketplace add openai/codex-plugin-cc then /plugin install codex@openai-codex and /reload-plugins.
Run /codex:setup — it will detect if Codex CLI is missing and offer to install it via npm automatically.
If not logged in, run !codex login to authenticate with your OpenAI account (free tier works).
Stage your changes in git, then run /codex:review --background to kick off a non-blocking review, check with /codex:status, and retrieve results with /codex:result.
For security/logic-sensitive code (e.g. Supabase RLS rules, auth middleware), use /codex:adversarial-review which actively challenges your implementation rather than just describing it.

How I could use this

Set up a pre-commit git hook in Henry's blog repo that automatically triggers /codex:adversarial-review on any changes to Supabase query files or API route handlers — surface the review output as a GitHub PR comment via a small Node script calling the Codex CLI directly.
Build a 'dual-AI review' feature into Henry's portfolio site: when he publishes a new blog post with embedded code snippets, a background job runs both Claude and Codex against the snippet and displays a side-by-side 'AI peer review' panel showing where they agree or disagree — a genuinely unique blog differentiator.
Use the /codex:rescue delegation pattern to build an async 'fix-it queue' for Henry's blog: when Lighthouse or Supabase logs surface an error, a webhook triggers a Codex rescue task in the background, and the result is posted to a private Slack/Discord channel with a ready-to-apply patch — turning monitoring alerts into draft PRs automatically.

6. sanbuphy/learn-coding-agent

11,331 stars this week · various

A reverse-engineered architectural breakdown of Claude Code's CLI agent internals — tool system, permission flows, sub-agents, and the 12 production harness mechanisms — compiled from public sources.

Use case

If you're building a coding agent or AI-powered CLI tool and don't want to reinvent the wheel, this repo maps exactly how a production-grade agent like Claude Code structures its query loop, manages 40+ tools with permission scoping, and layers safety/telemetry on top. Concrete example: you're building an AI blog assistant that can run code, fetch URLs, and edit files — this repo shows you how to architect the tool registry, permission gating, and sub-agent delegation so it doesn't become a tangled mess.

Why it's trending

Claude Code just crossed massive adoption numbers and developers are hungry to understand what's under the hood — this repo hit 11k stars in a week because it's the clearest public teardown of a real production agentic system at a time when everyone is trying to build their own.

How to use it

Clone the repo and read docs/en/ sequentially — start with the Architecture Overview section to understand the Entry → Query Engine → Tools/Services/State pipeline before touching any code.,2. Study the Tool System section to understand how Claude Code registers tools with permission scopes. Map this pattern to your own agent: define a ToolRegistry interface where each tool declares name, description, permissionLevel, and an execute(input, context) method.,3. Implement the core agent loop pattern extracted from the repo: while (!done) { const plan = await llm.plan(state); const tool = registry.get(plan.toolName); await permissions.check(tool, session); const result = await tool.execute(plan.args); state.update(result); } — this mirrors the query engine cycle described in the architecture.,4. Apply the '12 Progressive Harness Mechanisms' as a checklist for hardening your agent: rate limiting, sandboxing, telemetry, fallback modes, undercover/stealth mode for background tasks, and remote control hooks — treat each as a separate middleware layer.,5. Use the directory reference tree as a scaffold for your own project structure, mapping tools/, services/, and state/ as top-level modules in your Next.js API routes or a standalone Node.js agent process.

How I could use this

Build a 'Blog Writing Agent' as a documented side project: a CLI tool that takes a topic, uses Tavily/Exa to research, drafts a post via Claude, and auto-creates a Supabase record with draft status — then write a blog post on Henry's site specifically documenting how you architected the tool registry and permission system using lessons from this repo. Instant credibility content.
Create a 'Career Doc Agent' using this architecture: a multi-tool agent where Tool 1 scrapes a job description URL, Tool 2 diffs it against a stored resume in Supabase, Tool 3 generates a tailored cover letter, and Tool 4 saves the result — with permission gating so Tool 3 only runs after Tool 2 confirms a match score above a threshold. Mirrors the sub-agent delegation pattern in the repo.
Implement a lightweight version of the '12 harness mechanisms' as a reusable TypeScript AgentHarness class for Henry's AI blog features (e.g., auto-tagging posts, generating summaries on publish) — rate limiting LLM calls, logging tool usage to Supabase for observability, and a fallback mode that returns a cached result if the LLM errors. Open-source it as next-agent-harness and reference the Claude Code architecture as inspiration in the README.

7. ChinaSiro/claude-code-sourcemap

8,448 stars this week · TypeScript

Reconstructed TypeScript source of Anthropic's official Claude Code CLI (v2.1.88), extracted from npm sourcemaps — giving developers a rare look at how a production AI coding agent is actually architected.

Use case

Developers building AI coding tools have no reference for how to structure multi-agent coordination, tool dispatch (Bash, FileEdit, Grep), plugin systems, or voice interaction in a real production CLI. This repo lets you read exactly how Anthropic wired those pieces together — e.g., how coordinator/ manages parallel agent tasks or how tools/ implements safe file editing with rollback.

Why it's trending

Claude Code just had a major push as Anthropic's flagship agentic product, and this sourcemap extraction went viral on the Chinese dev community (linux.do) this week — it's effectively a leaked architecture tour of a $1B+ company's internal AI agent framework.

How to use it

Clone the repo: git clone https://github.com/ChinaSiro/claude-code-sourcemap && cd claude-code-sourcemap/restored-src/src,Browse the architecture starting points: main.tsx for CLI entry, coordinator/ for multi-agent orchestration patterns, tools/ for the 30+ tool implementations like BashTool and FileEditTool.,Study a specific pattern — e.g., read tools/FileEdit.ts to understand how they implement safe, atomic file mutations with diffing before applying changes to disk.,Cross-reference with the actual npm package to verify behavior: npm install @anthropic-ai/claude-code@2.1.88 and compare CLI behavior against what you read in the source.,Use the skills/ and plugins/ directories as a blueprint if you're building an extensible AI tool system — these show how Anthropic registered capabilities dynamically rather than hardcoding them.

How I could use this

Steal the coordinator/ multi-agent pattern to build a blog post draft pipeline: one agent researches a topic via web search, another writes the draft, a third critiques it — all coordinated like Claude Code's parallel task system, orchestrated from a single Next.js API route.
Use the commands/review.ts implementation as a reference to build a resume/cover-letter review tool in your career section — the pattern shows how to pass structured file context to Claude with a specific critique prompt template and stream the response back incrementally.
Clone the plugins/ registration pattern to make your blog's AI features extensible: define a PluginRegistry in TypeScript where each AI feature (summarizer, tag sugguster, reading-time estimator) is a self-contained plugin that gets loaded at runtime, so you can add new Supabase-backed AI tools without touching core blog logic.

8. Kuberwastaken/claurst

8,261 stars this week · Rust

A clean-room Rust reimplementation of Claude Code's terminal coding agent — with no telemetry, unlocked experimental features, and multi-provider support — built by reverse-engineering Claude Code's leaked spec.

Use case

Claude Code's official binary phones home and locks certain features behind Anthropic's infrastructure. Claurst solves this by giving you a fully local, auditable terminal agent that can edit files, run commands, and reason about your codebase — without any tracking or vendor lock-in. Concrete example: you can point it at Henry's Next.js blog repo and have it autonomously refactor components, write Supabase migrations, or generate TypeScript types — all from the terminal, with any LLM provider.

Why it's trending

Claude Code's source leaked this week, triggering a wave of community analysis and clean-room rewrites — this is the fastest and most complete Rust port to emerge from that moment. The timing coincides with developers actively looking for self-hostable alternatives to Anthropic's locked-down CLI.

How to use it

Clone and build: git clone https://github.com/Kuberwastaken/claurst && cd claurst && cargo build --release,2. Add the binary to your PATH: export PATH=$PATH:$(pwd)/target/release (or move claurst to /usr/local/bin),3. Run in your project root: claurst — it will drop you into an interactive terminal session aware of your current directory and file tree.,4. Connect your preferred provider: type /connect inside the session and follow the prompts to wire up OpenAI, Anthropic, or another compatible endpoint via API key.,5. Issue natural-language tasks directly: e.g. Add a Supabase RLS policy to the posts table so only the author can update their own rows — Claurst will read relevant files, propose diffs, and execute with your approval.

How I could use this

Wire Claurst into a local dev workflow script that watches Henry's /content directory for new MDX blog posts, then auto-runs Claurst to generate OG image alt text, extract tags, and create a Supabase insert migration — all as a pre-commit hook, zero UI needed.
Build a 'resume diff' CLI tool on top of Claurst: paste a job description as a prompt, point Claurst at a /resume markdown file, and have it output a scored gap analysis plus a rewritten bullet list tailored to that JD — useful as a career tool Henry could open-source and blog about.
Use Claurst as the backend 'brain' for an AI code-review feature on the blog itself: when Henry publishes a post with embedded code snippets, a GitHub Action triggers Claurst against those snippets to generate a 'senior dev notes' aside — surfacing potential bugs or improvements directly in the post as a collapsible MDX component.

9. titanwings/colleague-skill

7,798 stars this week · Python

colleague-skill converts a departing coworker's chat logs, docs, and emails into a deployable AI agent that mimics their working style, technical opinions, and communication patterns.

Use case

When a key engineer leaves, they take institutional knowledge with them — undocumented architectural decisions, personal coding conventions, and tribal context that no handoff doc captures. This tool ingests raw Slack exports, markdown files, or pasted chat history and synthesizes a Claude-powered 'skill' that answers questions the way that person would, flags the same edge cases they'd flag, and even knows when they'd typically pass the buck. Example: your lead backend dev quits, you feed in 6 months of Slack threads and their internal wiki pages, and the resulting skill can answer 'why did we choose Postgres over MySQL for this service?' in their voice.

Why it's trending

It went viral this week on Chinese dev Twitter/X as a darkly comedic response to mass tech layoffs and the AI-displacement anxiety cycle — the quote at the top ('you AI people are traitors, you already killed the frontend devs, now you're killing backend, QA, devops...') struck a nerve globally, making it simultaneously a satire of and a genuine tool for the AI job disruption era.

How to use it

Clone the repo and install deps: git clone https://github.com/titanwings/colleague-skill && cd colleague-skill && pip install -r requirements.txt
Export your data source — for Slack use the API token flow, for markdown/email just drop files in data/raw/
Run the ingestion pipeline: python ingest.py --source slack --token YOUR_TOKEN --user @alice (or --source markdown --path ./docs/alice/)
Generate the skill: python build_skill.py --name alice --description 'Senior backend, owns payments service, strong opinions on API versioning' — this produces a colleagues/alice.skill JSON file
Invoke it via Claude Code: load the skill file as context and query it — e.g. claude --skill colleagues/alice.skill 'How would Alice structure the retry logic for the webhook handler?'

How I could use this

Build a 'Blog DNA' skill from Henry's own published posts — feed in all existing markdown files to create a writing-style agent that drafts new posts in his voice, flags when a draft sounds too generic, and suggests phrasing he'd actually use. Wire it into a Next.js /api/draft-assist endpoint backed by Claude API.
Create a 'Past Henry' career agent from old resume versions, cover letters, and LinkedIn exports that answers recruiter screening questions, identifies skill gaps between his profile and a job description, and generates targeted cover letters — stored as versioned .skill files so he can track how his positioning evolves over time.
Add a 'Reader Memory' feature to the blog: after a visitor reads 3+ posts, synthesize a lightweight skill from the posts they engaged with and use it to power a 'you might think about...' sidebar — a Supabase edge function checks read history, Claude with the post-collection as context generates a personalized next-read recommendation that explains why in the reader's apparent interest framing.

10. emdash-cms/emdash

7,409 stars this week · TypeScript · astro cms emdash typescript

EmDash is a type-safe, serverless CMS built on Astro + Cloudflare that modernizes WordPress's extensibility model with sandboxed plugin isolation and zero PHP.

Use case

WordPress's plugin model is a security nightmare — one compromised plugin owns your entire server. EmDash solves this by running plugins in sandboxed Cloudflare Worker isolates, so a rogue plugin can't touch your DB or filesystem. Concretely: you can install a third-party comment moderation plugin without worrying it exfiltrates your content or API keys.

Why it's trending

It launched this week with a one-command scaffold and a direct Cloudflare deploy button, hitting the sweet spot of developer frustration with WordPress performance and Next.js CMS complexity simultaneously. The 'WordPress killer on serverless' positioning plus Astro's momentum is driving heavy sharing on X/dev Twitter.

How to use it

Scaffold a new project: npm create emdash@latest and pick the 'blog' template when prompted.
Configure your Cloudflare account (D1 database + R2 bucket) — credentials go in wrangler.jsonc. If you're on the free plan, comment out the worker_loaders block to skip sandboxed plugins.
Run locally with npm run dev — EmDash spins up an Astro dev server with the admin UI at /admin.
Deploy with npx wrangler deploy — your site, admin panel, and SQLite DB all land on Cloudflare's edge in one push.
Add content via the admin UI or write directly in Markdown — EmDash supports both a rich editor and file-based content.

How I could use this

Migrate Henry's Next.js/Supabase blog to EmDash as a side-by-side experiment — use EmDash's blog template as the public-facing content layer while keeping Supabase for user auth and AI feature data, connecting them via EmDash's plugin API to pull personalized post recommendations per logged-in reader.
Build a custom EmDash plugin (sandboxed Worker) that acts as a 'career tools' micro-CMS — Henry manages his resume bullet points, project descriptions, and skills as structured content in EmDash's admin, then the plugin exposes a typed JSON API that his separate cover letter generator and resume matcher apps consume, keeping all career content in one source of truth.
Write an EmDash plugin that intercepts the post-publish lifecycle event and calls an OpenAI endpoint to auto-generate: an SEO meta description, 3 tweet variants, and a TL;DR summary — storing them as post metadata fields accessible in the Astro template, so every published post instantly has AI-assisted distribution copy without leaving the admin UI.