docs(wiki): Agent in the Loop follows Require Tool Confirmation
Dangerous tools in a loop run now follow the project's switch. Turned off,
the run is unattended and the chat marks it with a red icon next to the
Agent in the Loop chip; the page says what that risks and that the vendors
expect an API key for unattended automation.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
docs(wiki): Agent in the Loop page, and links from the MCP, agent and provider pages
A new page for Agent in the Loop: how it works, setup for Claude Code and
Codex (a folder-scoped .codex/config.toml), what the drawer shows, human
approvals, Fireteam/LATS/Deep Think, several chats at once (one client or
several), limits, the security model, terms of use, troubleshooting, FAQ and
turning it off, with ten screenshots taken from the running app.
MCP-Server gains the three loop tools, the agent:loop permission and its rate
bucket, and states the MCP server is on by default for a local install; the
sidebar, Home, the AI Agent guide, AI Model Providers (not a provider),
Anatomy of a Think Step, Fireteam and LATS link to it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>