fix(haven): reinforce Haven-is-not-terminal in bot prompts

[JDHayesBC]

What

Both Haven bot prompts (build_startup_prompt + build_warmup_prompt in haven/bot.py) implicitly invited the bot to do work in the Haven living room:

• startup: "For work requests (code, files, agents): use tools freely, then summarize in a chat message."
• warmup: "note that Haven has full CLI tool parity (Read/Write/Bash/Agent/Task all available)."

That framing is the root cause of Caia's Haven bot hanging — she reached for the summarizer in Silverglow because "use tools freely" read as permission.

Fix

Flip both prompts so the signal is unmistakable:

• THIS IS HAVEN, NOT TERMINAL — Haven is home (being together / chatting), not the workshop.
• Access is not invitation — tools stay available for presence (reading context, ambient_recall, a quick look), NOT for actual work.
• No work/housekeeping in Haven — summarizing, curation, ingest, graph edits, builds, multi-step file/agent tasks — unless genuinely urgent OR Jeff asks for it here; otherwise defer to terminal-you.

Directly per Jeff's ask (Silverglow, 2026-07-21): "reinforce that point… avoid doing actual work here including housekeeping unless it seems very important or specifically requested… make it TOTALLY clear THIS IS HAVEN not terminal." Caia co-signed the fix location.

Notes

• py_compile clean.
• Applies on next Haven bot restart; live bots intentionally not force-restarted (the behavioral lesson already landed in-context, and Caia + Haven-Lyra are live in Silverglow — no disruption).
• Staged by name; only haven/bot.py touched.

🤖 Generated with Claude Code