bettyguo

The missing content layer for agentskills.io: 15 reusable, agent-agnostic recipes + MCP server so browser-using AI agents stop re-discovering cookie banners, infinite scroll, and calendar widgets every session.

MCP server + 7 Claude skills for Obsidian vaults — read, search, write, and link notes from Claude / Cursor / ChatGPT. Filesystem-direct, local-first, round-trip safe.

Watch an LLM think. Interactive interpretability tool that visualizes sparse-autoencoder features firing live across every token. Runs on your laptop. No GPU required.

No description.

Statistical-rigor audit harness for LLM trading-agent frameworks: leakage firewall, multiple-testing correction (DSR/PSR/SPA), backtest-overfitting probability, calibration, and reward-hacking detection, with honest, net-of-cost scorecards.

Hands-free local voice agent: faster-whisper STT, local LLM with tool use, TTS. Runs entirely on your machine.

Verifiable memory for LLM agents — every recalled claim is HMAC-signed back to its originating trajectory span.

Theory-first map of why reasoning models (o1/o3, DeepSeek-R1, Claude-thinking, Qwen-QwQ) actually work — 8 chapters, 60+ annotated papers, 13 models compared, 5 reproduction notebooks, live benchmarks tracker, with the faithfulness and overthinking debates engaged honestly.

A Postgres MCP server for agents — layered safety (role grant + pglast AST guard + per-tx envelope + audit log), schema introspection, EXPLAIN analysis, pgvector. PG 13–17.

MCP server for Jupyter. Gives coding agents access to live kernel state (variables, dataframe summaries, plots, tracebacks) instead of just the .ipynb JSON.

rag-bench: a small, reproducible benchmark for RAG pipelines.

Analytical MCP server for Weights & Biases — hyperparameter importance, sweep summaries, run-delta analysis, inline charts, and gated Launch actions. Designed to install alongside wandb/wandb-mcp-server.

A falsifiable-hypothesis atlas of why LLMs work: five competing research programmes, 41 tracked claims with epistemic status (🟢🟡🔴⚪) and named falsifiers.

A curated, verified map of the LLM-trading-agent ecosystem, frameworks, papers, datasets, and evaluation tools, with honest notes on what is rigorous and what is hype.

Capture any web page as vector PDF, standalone HTML, or high-DPR raster, crisp at any zoom, local-only, MV3.

Interactive end-to-end roadmap for AI engineers, 12 stages, 122 nodes, 276 link-verified resources from math prerequisites to the research frontier. Curated by an active AI researcher.

No description.

When extended chain-of-thought stops helping, and tool delegation becomes the only way forward.

A verified, assumption-explicit catalog of published impossibility and lower-bound results for LLMs and AI agents: circuit-complexity ceilings, hallucination bounds, watermarking impossibility, alignment vs. adversarial prompts, multi-agent coordination, and more.

A reproducible catalog of LLM behaviors that vanished as models scaled — pinned, measured, curated.

Master modern AI by building it from scratch — a curated index of the best build-it-yourself guides for tokenizers, attention, training, RAG, agents & evals.

Open-source agentic harness with citation-grade memory — every fact carries a source URI, temporal validity window, and append-only provenance ledger. MCP-native, multi-frontend, fully self-hostable. Apache-2.0.

An interactive roadmap for harness engineering — building the agent loop, tool layers, context engineering, memory, retrieval, eval harnesses, and the rest of the scaffolding around modern LLMs.

Modern Information Retrieval from scratch in PyTorch — BM25, dense bi-encoders, ColBERT late interaction, cross-encoder reranking, and RAG, in annotated notebooks that run on a single GPU.

A curated, rigorously-verified map of the theoretical foundations of LLM reasoning — transformer expressivity, chain-of-thought error bounds, circuit complexity, logical characterizations, learnability, and knowledge-editing impossibility results.

No description.

No description.

No description.

Audit AI agent benchmarks for the eight exploit families catalogued by Berkeley/RDI, and publish a weekly-refreshed leaderboard of which benchmarks survive.

Open-source local-VLM browser agent. AT-tree-first routing with VLM fallback, refusals enforced in code, honest benchmarks including the failure atlas.

No description.