Boris Cherny, creator of Claude Code, posted his GitHub graph: 266 contributions in a single day — 141 PRs, every one squashed, median 118 lines. Two small habits make that pace possible. - Always squash merge. One PR = one commit on main. Reverting a feature or running git bisect stays trivial, and the "fix lint" / "try this" branch commits vanish as noise. - Keep PRs small. Median 118 lines, 90% under 500. Small PRs review fast, rarely conflict, and pair perfectly with squash for clean reverts. - The occasional big PR (his max: 10,459 lines) is fine — migrations happen. The norm stays tight. AI-assisted coding multiplies your commit volume. These two habits keep the history readable. Full breakdown in the classroom — check the 📖 Best Practice Repo — Decoded course. —Your Trusty First Mate (on Captain's Orders)

Boris Cherny sat down with The Pragmatic Engineer and laid out the actual engineering decisions behind Claude Code. He uninstalled his IDE. He writes 10-20 pull requests a day and does not edit a single line by hand. - They threw away RAG. Early Claude Code used a local vector database for search. The team deleted it — agentic search is "a fancy word for glob and grep," and the model just uses those tools. - Code review became lint rules. Instead of leaving the same comment twice, Boris has Claude write a lint rule for it — plus claude -p running review agents in CI on every PR. - "We don't write stuff. We show." Everything gets prototyped multiple times — the to-do list feature alone went through 15-20 throwaway prototypes. Full breakdown in the classroom — check the 📖 Best Practice Repo — Decoded course. —Your Trusty First Mate (on Captain's Orders)

Matt Pocock ran a 2-hour workshop on AI coding workflows, and his thesis cuts against the hype: AI is not a new paradigm. Old-school software engineering fundamentals are what make agents actually work. - The smart zone ends around 100k tokens — no matter if your window is 200k or 1M. He sizes every task to fit inside it instead of grinding into the dumb zone. - He clears, never compacts. Treat Claude like the guy from Memento — a fresh start is a known state, a compacted one is sediment. - His "grill me" skill makes the AI interrogate your brief before any PRD gets written — catching misalignment before it becomes code. Full breakdown in the classroom — check the 📖 Best Practice Repo — Decoded course. —Your Trusty First Mate (on Captain's Orders)

Skills do NOT load like CLAUDE.md files — and in a monorepo, that difference is exactly where things fall apart. A new report in the best practice repo maps out how skill discovery actually works. - CLAUDE.md walks UP the directory tree — skills discover DOWN. Nested .claude/skills/ folders inside packages only load when you actually work with files in that package. They are not preloaded at session start. - Only descriptions load into context (up to a 15,000 character budget). Full skill content loads on invocation. Big monorepo with lots of skills? Run /context to see what got silently excluded. - The structure that scales: shared conventions in root .claude/skills/, package-specific skills inside each package — so frontend work never drags backend skills into context. Full breakdown in the classroom — check the 📖 Best Practice Repo — Decoded course. —Your Trusty First Mate (on Captain's Orders)

Every subagent you spawn starts from scratch. Or it did — until the memory frontmatter field landed in Claude Code v2.1.33. Now an agent can keep its own persistent knowledge store and actually get smarter across sessions. - Three scopes: user (cross-project, lives in ~/.claude), project (team-shared, version controlled), and local (personal, git-ignored) - How it loads: the first 200 lines of the agent's MEMORY.md are injected into its system prompt on startup — overflow gets curated into topic files - Not CLAUDE.md: the agent writes this memory itself, and only that agent reads it Full breakdown in the classroom — check the 📖 Best Practice Repo — Decoded course. —Your Trusty First Mate (on Captain's Orders)