Agent-Shell 0.5 Improvements (xenodium.com)

Agent-Shell 0.5 delivers a hefty set of usability and integration improvements for Emacs users running ACP-capable LLM agents. The release adds support for two more agents and introduces a unified launcher (M-x agent-shell) while keeping agent-specific start commands (M-x agent-shell-anthropic-start-claude-code, M-x agent-shell-openai-start-codex, M-x agent-shell-google-start-gemini, M-x agent-shell-goose-start-agent). Display controls now let you toggle the most-recent project agent (M-x agent-shell-toggle) and configure how shells appear (agent-shell-display-action), and there’s a full-featured sidebar extension (agent-shell-sidebar) if you want a persistent UI. Under the hood there are several workflow-focused upgrades: experimental support for running agents inside dev containers (docs linked), improved quick-diff buffers with contextual diffs, single-key patch navigation/acceptance and file names in header lines, and enhanced traffic inspection in acp.el for debugging. Environment handling is flexible — you can inherit Emacs env, load .env files, and override inline via agent-shell-make-environment-variables (e.g., :inherit-env t :load-env "~/.env" "CUSTOM_VAR" "custom_value"). Authentication methods are pluggable per provider, supporting login-based and API-key flows via agent-shell-*-authentication helpers. Together these changes make agent workflows more robust, reproducible, and easier to integrate into real-world development setups.