What to do

  1. If you sync continue, run agnostic-ai sync and commit the move of skills to .continue/skills/<name>/ (one sync and one commit).
  2. If you imported settings from several tools, run agnostic-ai import all again and delete the old .agnostic-ai/settings/imported.yaml (one file per project).
  3. If you use sync --global with a root variable such as CLAUDE_CONFIG_DIR, delete the files an earlier sync left under the old root (one directory per tool).

Nothing else needs action.

What changed

  • A settings spec's effort sets the default reasoning effort for Claude Code, Copilot, Codex, and Factory, each under its own key (#1069, #1089).
  • sync --global writes native agents for 18 targets and stops before overwriting a file you edited by hand; --backup keeps a copy (#1033, #1082).
  • import all keeps every tool's settings in its own settings/<target>.yaml instead of only the last tool's (#1096).

Everything else is in the GitHub release notes.

Upstream CLI and model news

  • VS Code (2026-09-16): the MCP page now lists a workspace-root .mcp.json with a top-level mcpServers object as a standard location, and trusted workspaces start its servers without a separate prompt. The claude target writes that file, so VS Code loads its servers too. State: supported. MCP servers
  • Cursor (docs checked 2026-09-23): subagents also load from .claude/agents/ and .codex/agents/, and .cursor/ wins a name clash, so a repository syncing claude and cursor has each agent in two places. State: supported. Subagents