Release v4.59.0

The Screenplay MCP server no longer needs a fixed folder, the documented growth prompts work, and the board follows the model on disk.

Added

• screenplay mcp can start without a root. It picks the model folder when it first needs one: a path given to open-workspace, then the single root the MCP client offers (re-read when the client changes its roots), then the folder it was launched from when that already holds .play files. If none applies it asks for a path instead of guessing.
• open-workspace accepts an optional path, which also switches a session to another folder.
• Open the board fullscreen and keep prompting: it checks the model on disk every couple of seconds while it is visible and redraws when a file changes, whether the assistant applied a change or you edited one by hand.
• Workspace readiness reports whether the source is accepted for authoring separately from whether it is in the executable subset, so a full-language model is no longer described as broken.

Changed

• apply and recover-workspace are no longer marked destructive, so clients that confirm every destructive call stop asking each time. Both stay journaled and recoverable.
• A rejected proposal now lists the diagnostics (code, location, message) instead of a generic compilation failure.

Fixed

• add-document no longer demands an internal stableKey; it is derived from the file name.
• diagnostics and describe-application answer for a new, empty folder instead of failing.
• The documented authoring flow no longer stops at an empty folder, a missing stableKey or an unexplained rejection.