optimize

To manage frontier state and candidate briefs, you coordinate algorithm-first quests by shaping briefs, ranking them on a single surface, and promoting the winner into a durable optimization line. This replaces ad-hoc exploration with a disciplined workflow.

The best way to direct algorithmic exploration toward a durable improvement path is to invoke the optimization stage, which manages a small within-line candidate pool, performs bounded smoke tests, and drives measured experiments toward a clear route.

You rank and promote candidate solutions by managing a small within-line candidate pool, performing bounded smoke tests, evaluating the frontier state, and promoting the highest-ranked candidate into a durable line.

Yes, this optimization workflow supports internal submodes including brief, rank, seed, loop, fusion, and debug, functioning together as a unified optimization workflow to evaluate and select optimal routes.

Yes, the optimization workflow specifies explicit frontmatter requirements, optional resources, and safety constraints, including explicit instructions for 'bash_exec' usage and artifact recording during execution.

You should use a structured optimization workflow instead of ad-hoc exploration when your algorithm-first quests require disciplined exploration, evaluation, and route selection to turn frontier insights into a single durable improvement path.