dream-num/univer-craft/univer-craft-yolo

Univer Craft YOLO

Use Univer-specific context to delegate the requested outcome to $tina-yolo. This mode depends on both $univer-craft and $tina-yolo; it does not maintain a second execution workflow.

Read the installed $univer-craft skill's Shared context and research and Repository defaults sections, including its SDK research map as relevant. Apply those sections without running its normal dispatch or next-step handoff. When installed together, the shared skill is at ../univer-craft/SKILL.md; otherwise locate it through available skills.

Read and invoke the installed $tina-yolo with the original request, target repository, acceptance criteria, existing Change/Proposal Plan paths, SDK layers and versions, relevant sources, and existing Research Notes. Pass unresolved SDK questions to its research stage; carry out bounded research as required by the shared guidance. Do not research every SDK or repeat sufficient research.

Invoking this mode delegates the requested task under Tina YOLO's authorization and routing rules. Apply its overrides to nested Tina stages and agents. Let the orchestrator decide scope details, architecture, UX, and engineering choices, recording assumptions and evidence in Tina's existing run record. New apps use pnpm and TypeScript by default; existing apps retain their conventions. Do not ask preference questions or stop with instructions for the user to invoke the next stage.

Tina YOLO owns planning, implementation, local commits, independent QA/review, verification, fix loops, and continuation. Preserve its completion criteria and permission boundaries; requesting this wrapper does not authorize pushing, deployment, destructive actions, or archive. On continuation, resume the existing run after checking its artifacts and evidence.

If a required SDK capability, license, credential, service, or Tina prerequisite is unavailable, record the exact blocker and finish independent work. Do not replace requested integration with a mock and declare completion. Return the delivered behavior, checks, Research Note/run paths, and any incomplete work in the user's language, without an intermediate stage handoff.