execute-story
Execute Story
Drive an approved story-unit plan to completion: derive review units from requirement stories, implement unit-by-unit with fresh subagents, stop for a human after each unit, resume from the unit ledger, finish with whole-branch review.
This is not continuous execution. Continuous multi-task orchestration
without unit barriers is execute-plan. Controller-implements-without-subagents
is execute-inline. If Execution-mode: is continuous, stop and use
execute-plan instead. If the user wants no subagents, name execute-inline.
Why fresh subagents: each worker gets only its task brief; bulk artifacts
travel as paths under .skills/, never pasted session history.
Narration: at most one short line between tool calls. Ledger + tool results carry the record.