the living project: memory as leverage

the session forgets; the project must not

Each session starts with a fresh context window (memory docs), and module 1 taught you to treat that as a feature: clean context, one task. The cost surfaces across weeks. A correction you gave in March gets re-explained in May, and the third session makes the mistake the first one already fixed. Compounding requires writing things down where the next session reads them.

You met the first store in module 1: CLAUDE.md, the rules you write. The docs add its counterpart, auto memory (notes the agent writes for itself across sessions: learnings, commands, and preferences, stored per repository and loaded at session start): "notes Claude writes itself based on your corrections and preferences", stored per repository, with an index loaded at every session start. You write the law; the agent keeps the diary.