What each release since 1.0.0 changed, newest first. Every entry is a summary of the README's Upgrading, which says, release by release, what a workspace sees; 1.3.1, a documentation-only patch with no section there, is summarised from its GitHub Release.
status --json gains forge.newerTag, always present and last in forge: the remote Forge's highest vX.Y.Z tag above a workspace ref that is itself a vX.Y.Z tag, else null. The status tool of craftar mcp and craftar ui's workspace answer carry it too.status gains one line when there is such a tag. No byte sync writes, no lock, no exit code and no network call changes: the tags come from the fetch status already makes, or from the cached copy.agents-md "no AGENTS.md equivalent — skipped" warning is no longer printed for an ingredient another target of the workspace writes; it still names every ingredient no other target writes. doctor --strict can now exit 0 where that warning was its only warn.craftar drift promote: a new variant whose base a recipe in the profile's own folder cites no longer refuses: when the profile resolves that recipe, it is repointed to the variant behind the same git gate and proof.opencode: a remote MCP server whose headers is not an object is dropped with a warning naming the server. craftar setup: a new Forge's README advises keeping the Forge out of a workspace's git repository.craftar setup: creates a Forge from scratch (or by importing a workspace's .claude/), makes it a git repository with one initial commit, and starts workspaces on it. It is the one craftar command that commits, only in the directory it creates or finds empty, and it never pushes. See Setup.craftar with no arguments, in a terminal, on a machine with no registered workspace, ends its help with new here? run craftar setup; the interactive craftar init suggests it when a Forge answer names no Forge.opencode (OpenCode 2 or later): agents to .opencode/agents/<name>.md with permissions derived from their tools, commands to .opencode/commands/<name>.md, MCP servers to .opencode/opencode.json; rules reach OpenCode through agents-md.opencode needs schema: 18; craftar.lock is written at schema 7 when it names it.opencode sees no change.cursor, copilot, agent-skills, codex and antigravity were checked by opening each tool from its CLI on a synthetic workspace; the README says what was validated, per target.copilot also writes .github/mcp.json for the Copilot CLI; claude-code leaves out an empty tools line in an agent.targets.sync removes them, and keeps a hand-edited one as orphan-drift. antigravity covers the same use.registration is written for Codex to .codex/hooks.json and for Antigravity to .agents/hooks.json.cursor and copilot run the hooks .claude/settings.json holds, so no file of theirs is written.codex: agents to .codex/agents/<name>.toml; rules through agents-md, commands and skills through agent-skills.antigravity: rules to .agents/rules/, agents to .agents/agents/, MCP servers to .agents/mcp_config.json.schema: 16; a workspace that names neither sees no change.cursor (.cursor/rules/, .cursor/agents/, .cursor/mcp.json), copilot (.github/instructions/, .github/copilot-instructions.md, .github/agents/, .github/prompts/, .vscode/mcp.json) and agent-skills (.agents/skills/).schema: 15; a workspace that names none sees no change.--json shape, the loading of a Forge schema or the reading of the lock needs a major version, and a change to generated bytes comes in a minor.Before 1.0.0: the 0.x history, release by release, is in the README's Upgrading and on GitHub Releases.