End to end: plan → implement → PR on a real feature

So far every lesson focused on one piece: CLAUDE.md, a good prompt, plan mode, permission modes, subagents, MCP, hooks, git, PR, code review, self-review. This lesson doesn't teach a single new idea — it connects all of them into one sequence, on a small, real feature in the driftlog-cli project: re

Working with an agent on a whole feature, from start to finish, is like cooking a meal in a kitchen that isn't yours: you check what's in the drawers first, plan before turning on the stove, and taste it yourself before serving it to guests.

end-to-end workflow: The full sequence of steps in working with an agent on one feature: reading context, prompting, planning, choosing permissions, implementing, self-reviewing, committing, and opening a PR — not each step alone, but the order and the choices that connect them.
scope-matched diff: A diff where exactly the files the plan promised changed, and no more — the sign that a pre-commit self-review came out clean.
draft-first PR: Opening the PR as a draft right after push, and marking it ready for review only once CI is green — so review isn't invited before the code has even passed automated checks.