Now I'll build this as a rich, interactive HTML playbook artifact.Here's your complete **Claude Code Session Discipline Playbook** — interactive, tabbed, and built for your exact stack. Here's what each pillar gives you: **P1 — Scope:** A decision tree for every file-read question, the 4-file ceiling visualized, and two concrete bad→good scope examples pulled from your actual service/Prisma/test structure. **P2 — File Limit:** The definition of "large" tuned to this monorepo (test files always count, regardless of line count), plus a copy-paste `/read` command sequence for a real auth debugging scenario. **P3 — /compact:** The 60% threshold explained for your specific project (TypeScript interface chains + Prisma schemas bloat context faster than most), visual warning signals, and the exact 3-turn pre-compact ritual. The "never compact during" list is there to protect you mid-migration. **P4 — CLAUDE.md:** A ready-to-paste template skeleton plus the before/after comparison showing 127 tokens → 18 tokens on a single entry. The skills pattern keeps your base context clean every session. **P5 — Checkpoints:** Hook config for `.claude/settings.json`, the ASCII checkpoint template, and the 3 health-check signals that tell you your checkpoint has become a liability. The **Quick Ref** tab is the desk-pinnable card — all five rules, under 200 words, zero fluff.