Design-flow now sees leftover catalog skills and aesthetics-card commitments; reply-format injects present-shells/, and world-simulator prompts add random-range plus a tighter module set.
Co-authored-by: Cursor <cursoragent@cursor.com>
Narrator uses capability-probed structured delivery and falls back to shell body on failure; play UI hides worker questions and keeps thinking on process rows.
Co-authored-by: Cursor <cursoragent@cursor.com>
Opening accept no longer locks the graph; session.json stays creation-only while play starts from an instance snapshot. Sidebar products nest run saves, and play reading fills the feed at a denser type size.
Co-authored-by: Cursor <cursoragent@cursor.com>
Probe API profiles for json_schema/tools, flatten worker artifacts onto a strict schema, and let users spawn or revise graph nodes instead of a one-shot DAG.
Co-authored-by: Cursor <cursoragent@cursor.com>
Auditor packets can request generation-rule pools without LLM randomness; results land in 运行.本轮.旁观.生成抽样 and clear each play turn.
Co-authored-by: Cursor <cursoragent@cursor.com>
Main world layer can loop on the chance tool (requests array) until settlement JSON. Driver forwards output deltas; standalone chance worker accepts batch requests. play_slots.chance gates the harness.
Co-authored-by: Cursor <cursoragent@cursor.com>
Main world layer runs before narrator and deferred maintenance. Round-scoped settlement/maintain tags are archived at turn start so prior auditor table merges feed the next gm read.
Co-authored-by: Cursor <cursoragent@cursor.com>