Chat: markdown rendered by render-markdown.nvim (chat buffer only, not
markdown files), `## You` / `## Claude Code` headers as tinted bars,
upright code, no line numbers. Winbar shows icon, title and agent ·
model · effort, with a spinner and timer while the agent works; a lualine
extension shows context items and tokens. Highlights live in
lua/ai_ui.lua and follow gruvbox's light/dark switch.
Inline: the progress float is now a pill with per-part colours that
shows done/failed/cancelled for a moment; the diff banner is a label.
Inline edits always open the diff for review again (approval_mode auto
was applying them straight away), and run at medium effort with every
agent and model unless an adapter sets its own thought_level.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude Code and Antigravity chats open in the agent's own auto mode,
with CodeCompanion's approval mode on auto too. The CLI terminals run
`claude --permission-mode auto` and `agy --mode auto`. Inline stays as
is: it never prompts, and the agent's permission requests are what keep
it editing only the buffer copy.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The `ga` picker now lists only Claude Code and Antigravity. Every
built-in HTTP adapter is hidden after setup, so new presets in later
releases stay hidden too; hidden adapters still work by name.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Chat defaults to Claude Code (via claude-agent-acp), inline edits to
Antigravity pinned to Gemini 4 Hydrogen at high effort. The chat winbar
shows agent/model/effort, `ge` switches effort, and inline edits get a
progress float. Keymaps live under <leader>a.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>