Software Engineer
Things I use, things I recommend, things that save time.
How to stop getting rate-limited after an hour on Claude Code, Codex, or Copilot. Context window mechanics, the /effort command, model routing, kicking new conversations, and the silent token drains most people miss.
expect-cli reads your git diff, generates a test plan via AI, and executes it in a real browser with Playwright. Extracts cookies from your local Chrome/Firefox for authenticated testing. The validate step for agent loops.
qmd is a CLI semantic search engine for markdown by Tobias Lutke (Shopify founder). Hybrid BM25 + vector + LLM reranking. Local SQLite index, no cloud, no Docker. Pair with nano-graphrag for local graph RAG.
Google Stitch is an AI-native design tool from Google Labs that generates high-fidelity UI from prompts and exports real code. DESIGN.md encodes your design system as portable markdown any AI coding tool can read.
PinchTab is a 12MB Go binary that gives AI agents browser control via REST API. Serves the Accessibility Tree instead of raw HTML, cutting tokens by up to 90%. Stable element refs, stealth mode, zero Node/Python dependencies.
debug-bridge connects AI agents to your already-running webapp via WebSocket. Zero browser startup. Real-time DOM snapshots, console logs, and network visibility without Playwright cold starts.
agents-in-a-box is a Rust TUI with 71 skills, 37 agents, and a knowledge system that deploys to Claude Code, Codex, Copilot, Gemini CLI, Cursor, and 5 more platforms. Context engineering as a portable artifact.
The security agent stack behind Zerocool. Shannon for white-box pentesting (96% XBOW), PentAGI for autonomous scanning, Ghost Security for code analysis, Argus and WebCopilot for recon. All coordinated by one wololo agent.
code-review-graph uses Tree-sitter to build a local knowledge graph of your codebase so AI assistants only read files in the blast radius of a change. Claims 6.8x fewer tokens on code reviews.
OpenFang is a Rust-built agent operating system. 137k lines, single 32MB binary, zero dependencies. Seven autonomous Hands, 40 channel adapters, 38 tools, 16 security systems including WASM sandbox and Ed25519 signing.
here.now is free zero-config static hosting built for AI agents. Publish HTML, dashboards, and prototypes to a live URL in seconds via API. No signup, no deploy pipeline. Served from Cloudflare edge globally.
notebooklm-py and the NotebookLM skill give Claude Code source-grounded, citation-backed answers from Google NotebookLM. Zero hallucinations on your own docs. Programmatic access to notebooks, audio overviews, and quizzes.
Entire captures AI coding session transcripts in git commits and provides line-level AI-vs-human code attribution. Founded by ex-GitHub CEO Thomas Dohmke with $60M seed. Works with Claude Code, Codex, Gemini CLI.
Three voice coding tools compared. justspeaktoit (free, Deepgram, macOS menu bar), Parler (offline Whisper, cross-platform), and Wispr Flow (polished, paid). Talk to your coding agent instead of typing.
Every Code is the community fork of OpenAI Codex CLI with multi-agent orchestration, browser integration via CDP, auto-drive mode, and Claude/Gemini support. 3.7k stars, 9.4k commits. What Codex should have been.
OpenProse is a markdown-based programming language for long-running AI sessions. Contract-based semantics (requires/ensures), parallel execution, composable workflows. Thesis: an AI session is a Turing-complete computer.
Every approach to running AI coding agents remotely. DIY tmux+Tailscale, vibetunnel browser proxy, conductor.build, ClaudeHub, MobileClaude. Rated from free terminal hacks to dedicated mobile platforms.
Lightpanda is a Zig-based headless browser engine built from scratch for AI agent automation. No GPU compositor, no extension runtime. Purpose-built for fast headless testing without Chromium overhead. 13k+ stars.
Clawdbot (now OpenClaw) is a Chrome extension that relays browser control between AI agents and your real browser session. No headless browsers. Your cookies, your logins, your extensions. 247k GitHub stars.
OpenScreen is a free open-source screen recorder with auto zoom, cursor tracking, motion blur, and annotations. Does what Screen Studio does for $89 but costs nothing. macOS, Windows, Linux. 25k stars.
claude-phone by NetworkChuck gives Claude Code a real phone number via SIP and ElevenLabs voice. Call your agent, get call-back alerts. Impressive demo but STT-to-model-to-TTS latency makes daily use impractical.
dev-browser is a Claude Code skill that gives agents a sandboxed browser via QuickJS WASM. Persistent page state, full Playwright Page API, 30% faster and 40% cheaper than Playwright MCP for agent dev loops.
Obsidian Skills by Steph Ango (Obsidian CEO) gives Claude Code full read-write access to your Obsidian vault. Search, create, link, and organise notes through your coding agent. 22k stars.
GSD (Get Shit Done) is a meta-prompting framework with 69 commands and 24 agents that fixes context rot in long AI coding sessions. Phase-based workflow: discuss, plan, execute, verify. 48K GitHub stars.
Impeccable is a skill that installs frontend design taste into Claude Code, Cursor, and other AI tools. 20 slash commands for typography, colour, layout, motion. Curated anti-pattern rules stop AI defaulting to Inter font and purple gradients.
UI/UX Pro Max is a skill that gives AI coding agents design taste. 50+ UI styles, 97 colour palettes, 57 font pairings, 25 chart types, and design system generation. 60k+ stars. Stops agents defaulting to generic SaaS templates.
Stagehand adds three AI primitives to Playwright: act, extract, and agent. v3 is CDP-native (44% faster) with a caching system that replays actions at sub-100ms and zero LLM cost after first run. By Browserbase.
SuperClaude is a CLAUDE.md config framework with 30 slash commands, 20 agents, and 7 behaviour modes. Turns vanilla Claude Code into an opinionated development platform. 22k stars, 1.9k forks.
Scrapling is a Python scraping library with three fetcher tiers (HTTP, stealth browser, full Playwright), adaptive selectors that survive site redesigns, and Cloudflare bypass. 35k stars.
skills.sh is a package manager for agent skills. npx skills find to discover, npx skills add to install. Community registry with Vercel Labs, Browserbase, and community-published skills for Claude Code, Codex, and others.
ntfy and PingMe send push notifications when your AI coding agent finishes. ntfy is self-hostable HTTP pub/sub (one curl command). PingMe fans out to Slack, Telegram, Discord, Teams, and 10+ more from a single binary.
Claude Squad manages multiple AI coding agents in parallel using isolated git worktrees. Go-based TUI showing all agents at once. Works with Claude Code, Codex, Aider, Gemini. Zero merge conflicts. 6.8k stars.
StenoAI is a local-only meeting transcription app using whisper.cpp and Ollama. Records, transcribes, summarises, and lets you query meetings. Nothing leaves your machine. System audio capture, 99 languages.
markdown.new and Jina Reader convert web pages to clean markdown, cutting token usage by 80%+. markdown.new runs on Cloudflare edge (500 req/day free). Jina Reader uses a URL prefix for instant conversion including PDFs.
Claude Code Router is a middleware proxy that routes Claude Code API requests to DeepSeek, Gemini, Ollama, or OpenRouter. Same Claude Code UX, different models, lower bills. Rule-based routing by task type. 31k stars.
Deepgram Nova-3 provides real-time streaming speech-to-text fast enough for voice-driven coding. $200 free credit (12,000 minutes). Handles technical vocabulary like useState, async await, and middleware.
Claude Flow was the original multi-agent swarm orchestrator for Claude Code by rUv. SPARC methodology, 60+ agents, hive-mind architecture. Later rewritten in Rust as Ruflo with persistent memory and MCP bridging.
ccusage parses your local Claude Code session JSONL files and shows exactly where your tokens went. Daily breakdowns, per-model splits, 5-hour billing window tracking. Free CLI, no API keys needed.
markitdown by Microsoft converts PDF, DOCX, PPTX, images, audio, and HTML into clean markdown for LLM context. One CLI command, 95k stars. Supports OCR via LLM and audio transcription.
browser-use is a Python framework that gives LLMs agentic browser control via a DOM-first perceive-act loop. Works with Claude, GPT-4, Gemini, and local models. 89% on WebVoyager benchmark. 85k+ GitHub stars.