Garry Tan's thirteen opinionated workflow skills for Claude Code
gstack tool adding thirteen specialized workflow skills to Claude Code as slash commands (plan review, code review, QA, automation) turning one assistant into a team.
gstack tool adding thirteen specialized workflow skills to Claude Code as slash commands (plan review, code review, QA, automation) turning one assistant into a team.
MCP-based tool providing persistent cross-session memory for AI coding agents, storing decisions, preferences, and context across projects.
Tool for deploying autonomous AI agents on Amazon Lightsail. Limited content provided.
Concept article on local, secure AI architecture without cloud dependency. Limited detail provided.
CLI tool for managing AI-generated files in private Git-backed vault with symlinks, keeping them version-controlled but separate from public repos.
Intent: Developer tool tracking task goals and decisions in code repos for AI coding agents. Structured JSON metadata for decision rationale.
Votal AI open-source white-box red teaming tool for AI agents. Tests tool use, decision chains, real actions vs black-box approaches.
Empirical analysis testing whether LLM-generated synthetic data follows Benford's Law distribution patterns.
Largest AI Pokémon tournament as open benchmark for agent capabilities. Novel evaluation dataset and methodology.
Software architecture philosophy discussing human cognitive limits and LLM code generation. Design theory without concrete examples.
OpenShell provides private runtime environment for autonomous AI agents. Limited detail available.
Local SQLite-based memory layer for LLMs (Claude, OpenAI, Ollama) that persists context across sessions via prompt injection.
Discussion of technical interview challenges in LLM era. Companies struggling to differentiate between strong prompting skills and genuine architectural understanding.
Technical article on recursive language models for long-input reasoning, distinguishing them from standard long-context, retrieval, and agentic approaches.
One Man Company Talent Market is a self-hosted/remote AI agent marketplace with 160+ specialized agents across roles like engineer, designer, and researcher.
Metrx optimizes AI agents for ROI and revenue metrics rather than cost alone. Limited detail provided.
Forge DevKit enforces architecture patterns for AI coding agents via generated artifacts. Detects agent rationalization failures and provides intent routing for natural language commands.
FireClaw is an open-source security proxy that prevents prompt injection attacks on AI agents by filtering malicious web content through a 4-stage pipeline before agents process it.
Discussion of autonomous machine learning research and AutoResearch project direction.
Keypo: Secure Enclave encrypted secrets management system designed for AI coding agents.
Open-source meeting copilot with local transcription capabilities, no cloud dependency.
Tool to visualize and test AI agent behavior against GitHub repositories before execution on local systems.
Instar: persistent server wrapper for Claude Code agent with Telegram integration and evolutionary capabilities.
Mathematical proof demonstrating that more noisy features can outperform fewer clean features in machine learning models.
GPU Lite is minimal CUDA wrapper reducing bloat and complexity for GPU programming. Limited details provided.
Promptcmd: CLI tool enabling LLM prompts as executable programs with argument passing, piping, and SSH forwarding support.
AIBuildAI: AI agent that automatically builds and trains ML models. Ranked #1 on MLE-Bench benchmark.
Google DeepMind introduces framework to measure AGI progress using cognitive science taxonomy; launches Kaggle hackathon for evaluation design.
f0lkl0r3.dev: Searchable, interconnected database of 1,000 Computer History Museum oral histories using Gemini APIs for text processing and knowledge extraction.
Sashiko: AI agent system for automating Linux kernel code changes and development tasks.
Reticle: developer tool for testing LLM agent workflows. Integrates prompt testing, variable management, and tool execution in single interface.
ClawBlink enables creation of background AI agents triggered through Telegram or WhatsApp messages for automated task execution.
HiddenMRR uses GPT-4o to scan abandoned GitHub repos and identify B2B revenue opportunities from legacy code logic and market demand analysis.
Discussion comparing MCP (Model Context Protocol) and CLI approaches—headline only, no detail.
Unsloth Studio is an open-source platform with chat UI, finetuning for audio/vision/LLMs, auto data prep, code execution, and multi-platform support.
Cuckoo-GPU is CUDA implementation of Cuckoo Filters achieving 350x speedup. Lock-free GPU-accelerated data structure for high-throughput batch operations with configurable false positive rates.
Forge is an open-source governance plugin for Claude Code that manages 22 agents with quality gates and specification-driven development workflows.
PUNK is a tool for remotely controlling Claude Code running locally on a laptop, preserving local state while enabling remote access from mobile devices.
Llamactl: Self-hosted LLM manager supporting llama.cpp, MLX, and vLLM backends with web dashboard and OpenAI-compatible API.
SQLite-vector extension for approximate nearest neighbor search using HNSW indexing. Supports ACID transactions and multiple distance metrics for embedded vector search.
Antfly: Go-based distributed document database with full-text, vector, and graph search. Includes native ML inference via Termite service for local deployments.
Tool that indexes and ranks 58K AI agents with trust scores, enabling discovery and evaluation within the emerging agent ecosystem.
DashClaw: middleware intercepting AI agent actions before execution. Policy-based guard rails with YAML configuration for agent action control.
Research on automating forecasting question generation for AI evaluation. Evaluation methodology work.
Production architecture patterns for autonomous AI pipelines after 3 months deployment. Technical design insights for agent systems.
Hubcap: single Go binary wrapping Chrome DevTools Protocol in 118 shell commands. Designed to enable AI agents browser automation with JSON output.
Lewis 1.0: 8B parameter model trained on AI social data. Claims performance improvements on personality metrics versus Sonnet.
Praxis: AI-native intermediate language for agentic workflows. 51-token symbolic language converting English goals to structured auditable programs.
OpenReview: open-source self-hosted AI code review bot for GitHub PRs. Deploy to Vercel, powered by Claude, creates automated PR reviews.
MCP server enabling AI agents to upload local files to file.kiwi for instant shareable links. No auth required, supports large files.