LAB

Lab

Lectures are what I've thought through, case studies are what I've taken apart, and the Lab is what I've actually built — this section lists projects I made myself: most are open source, some run in production. They're all tools made for the same goal: keeping Agents from drifting on long tasks, giving them memory, keeping context under control, and making usage legible. One page per project: what it solves, how it works, how to use it.

Posts

LAB · 1 Open Source

harness-kit

The engineering method for "running long tasks with Claude Code", distilled into a project skeleton you can clone and use.

2 min read LAB · 2 Open Source

agent-memory-kit

A runtime memory layer for the product Agent you're building — four roles that distill experience into reusable lessons.

2 min read LAB · 3 Open Source

context-engineering-kit

One CONTEXT.md to manage everything that enters the context window across seven layers — each with a budget and a caching strategy.

2 min read LAB · 4 Open Source

claude-usage

One command reads your local Claude Code logs and works out tokens, cost, model split, and cache savings.

1 min read LAB · 5 Open Source

agent-skill-case-studies

Take well-designed Claude Agent Skills apart as teaching material and distill reusable design patterns.

1 min read LAB · 6 Open Source

superagent-from-scratch

An under-2,000-line, zero-framework, teaching-first rebuild of a modern SuperAgent harness — the agent core distilled out of deer-flow's 185k lines, built slice by slice in public, now at slice 11.

2 min read LAB · 7 Open Source

skill-catalog-pipeline

An engineering blueprint for turning 77k third-party skill packs into a semantically searchable (500ms) skill catalog foundation — a five-layer pipeline plus 17 hard-earned lessons.

1 min read LAB · 8 Open Source

meet-scribe

Automatic minutes for offline meetings — record → local transcription + speaker diarization → claude-generated minutes → Obsidian, fully unattended.

2 min read LAB · 9 Open Source

openknowledge-py

A from-scratch Python reproduction of inkeep/open-knowledge — a git-native markdown knowledge base for humans and AI agents. Understanding by rebuilding.

1 min read LAB · 10 In Production

Blue-collar Recruiting Voice Agent

An end-to-end phone pipeline — realtime voice conversations, call persistence, and an LLM grading loop — running in a real recruiting business.

2 min read LAB · 11 In Production

Delphi Clone · Digital Twin Platform

Distill a real person into a conversational digital twin — multi-source ingestion, adaptive interview drilling, grounded RAG answers, an alignment flywheel, and the person's own voice — a full pipeline built from scratch and running in production.

3 min read