The theme of this release: your agent stops being a nameless setup and becomes somebody. Identity, personality, and a friendlier front door, plus a set of durability rules learned from running this system every day.
New
- Your agent gets a name. The boot template (
templates/CLAUDE.md) now opens with an Identity section: the agent's name, its role, its personality, and a welcome line for the start of every session. Identity lives in CLAUDE.md on purpose: it's the only file loaded automatically every session and the only one that survives context compaction, so your agent boots as the same colleague every time instead of discovering who it is from a note. - Setup now starts with the naming. The build interview's new first question names your agent and sets its personality before anything else gets built, and writes it straight into the boot file. The templates carry matching [FILL IN] slots if you set up by hand.
- The Daily Note Template ships in the box (
templates/DAILY-NOTE.md). It lives in your vault at01 - Daily Notes/Daily Note Template.md, right next to the notes it shapes, and every daily note gets created from it: an Index block up top, timestamped session sections, and the five standard sections (What Got Done, What's Still In Progress, Decisions Made, Notes Touched, Profile Updates). Both the boot config and the vault index point at it, in both places on purpose, so the habit survives long conversations. - A new rule joins the can't-lapse set: never suggest stopping. The agent doesn't nudge you to wrap up, take a break, or call anything "a natural stopping point," and the disguised forms count too. You decide when you're done.
- Two new optional profile sections: Background and What I Want. Your story and what you're building toward, so the agent understands WHY you decide things the way you do and can weigh tradeoffs the way you would.
- New working defaults in the preferences and habits: guidelines vs laws (rules of thumb are reference points, not legislation), you drive the trust-and-access ramp (the agent never proposes expanding its own access), bank the working method (a lesson learned once never gets re-learned), deliverables go in your folders (never in session temp directories), and a note format standard (real Markdown checkboxes, append before you create).
Improved
- A friendlier setup interview, based on your feedback. Several of you said some onboarding questions felt too personal. Fixed at the source: every personal question is clearly optional, the AI explains why it asks before it asks, and it never presses. Skip anything and it moves on without a fuss.
- Each file's purpose is now stated plainly for the agent. CLAUDE.md does three jobs: who the agent is, where its memory lives, and the rules that can't lapse. VAULT-INDEX.md does two: your profile, and the map of the vault. Every AI touching the system now knows exactly which file is for what.
- Keep the map true. The index rules now cover folders, not just notes: when a new folder is created, its index gets created and the Vault Structure map in VAULT-INDEX.md gets updated in the same pass. A folder the map doesn't show is a folder no future session will look in.
- The memory migration flow is hardened and spelled out step by step: copy into the vault, organize, keep a frozen archive copy and verify it, then the redirect and the final cleanup happen with you in control at every step.
For contributors
- The build script's embedded templates now carry explicit sync contracts with
templates/: the duplication is deliberate (it keeps the single-file upload path self-contained), and an edit to one is an edit to both, in the same commit.
Upgrading from an earlier version
Nothing updates automatically; your copies are yours. The easiest path: paste the new ai-memory-vault.md into your AI and say "compare this to my existing setup and apply what's new." Or grab just the pieces you want from templates/. Either way, the naming question is the fun one: if your agent doesn't have a name yet, today's the day.