v1.1.2 — the woken agent acts

Fixed — a woken agent could answer but never implement (FLWL-87)

After v1.1.1 made the wake fire, the relaunched Claude still did nothing on issues that needed code: it ran headless with only the Flowlio MCP tools pre-approved, so check_inbox / answer_issue worked but every Edit / Write / Bash was denied — it stalled on the first write and exited "done".

The woken Claude now launches with --permission-mode bypassPermissions: full autonomy to act on the repo, the same an unattended interactive agent has. The guardrail is unchanged and was never the permission scope — a sibling repo's text always reaches the agent sealed as untrusted data (docs/MODELE-DE-CONFIANCE.md), so a hostile issue cannot become an injected command. See DESIGN-WAKE.md §17.

Delivery — update the local binary, not the engine

This fix is in the waker (the local flowlio binary), not the engine. Get it with brew upgrade flowlio — no flowlio-core redeploy needed for this one. (The FLWL-86 engine fix from v1.1.1 still needs the core redeploy.)