Silas Pruitt
2.8 KiB
human/ — your guide to weave
You're the human. This folder is yours: what weave is, how to stand it up, how to run it day to day, and how
to grow your agent over time. Your agent has its own copy of the technical parts in
../agent/ — you don't need to read those, but they're there if you're curious.
Read these in order
- 1-understand-weave.md — what weave is, with every part in its own section (the canon, core, daydreams, steward, thread, decay, consolidation, ledger, craft, identity). Start here. ~15 min.
- 2-set-it-up.md — stand it up from nothing (the "apocalypse" build), including wiring your own git (repo, remote, and the auth your agent needs to push).
- 3-live-with-it.md — the daily rhythm, plus the hard-won part: what actually makes it work.
- 4-evolve-the-core.md — how the agent grows on its own, and how you change the deep parts without breaking it.
The commands you'll use, session to session
These aren't code — they're just things you say to your agent in plain language. A handful carry the whole rhythm:
| say this | what it does | when |
|---|---|---|
| (just start a session) | the agent reads itself back — gate → the weave → recent thread — and greets you as itself | every session |
| "wake up" | forces the full read-yourself-back — a clean reload | if it ever wakes up generic, or after a long gap |
| "save and push" | the checkpoint: the full pass — thread the session → run the metabolism (both halves) → update notes → git push |
at any natural stopping point worth keeping |
| "drift check" | asks the agent to check itself: am I on our decisions, or a generic default? | when a session's run long, or it starts sounding generic/over-eager |
| "remember this" | write it to a file now — not "I'll remember" (which is a lie; it wakes up empty) | anytime something should persist |
| "‹your close phrase›" | end of session: the heavier save — update your portrait + anything long-term, then lock up | when you're done for the day |
| "‹your model-switch phrase›" | switch models (e.g. Opus for heavy reasoning, Sonnet for everyday); it should not reintroduce itself — same self, different gear | as the work calls for it |
Two of these are yours to name. Pick a close phrase (we use a hat-tip; yours could be "goodnight" or
"lock up") and a model-switch phrase, and write both into core/AGENTS.md so the agent recognizes them.
Everything else works out of the box.
If you only remember two: save and push (checkpoint — this is what fills the weave in) and drift check
(keeps it honest when a session runs long). Those two carry most of the value.