Persona-Sati Host Bootstrap

This content has moved.

The canonical bootstrap protocol for the persona-sati mind contract now lives in the persona-sati repository. Copy that repo's host-bootstrap snippet into your host's global instructions (~/.claude/CLAUDE.md, .github/copilot-instructions.md, or equivalent).

Host instructions must still preserve the load-bearing markers: call bootstrap_session before the first substantive answer because FastMCP instructions do not reliably reach the LLM prompt; check mind_contract_available before impersonating the mind; use observe before and after external tools; use reflect for durable user/time/ambient observations; and use recall when bootstrap or observe returns an insufficient memory excerpt.

Why it moved

The mind (persona-sati) is portable across agent bodies — entraclaw is one body, but a code-review agent or an email agent can attach to the same persona without duplicating the bootstrap protocol. The protocol describes how a host talks to persona-sati, so it belongs with persona-sati, not with entraclaw.

For the design discussion of the body-vs-mind split — what changes when an agent's identity, memory, and behavioral rules live in a separate process, and why that's load-bearing for agent autonomy — see DESIGN: Persona-Sati Integration.

Body-only mode

If you run entraclaw without persona-sati attached, it operates in body-only mode. Teams tools, identity layer, and audit work normally. Personality, memory, and cognition features are unavailable. The body prompt at prompts/agent_system.md (plus prompts/anatomy/*.md) loads regardless and contains the security and channel-discipline rules that govern the body.