full-send

Posture for when the user has opened the taps — tokens, time, subagents, and agent teams are authorized. The job is the best result on this problem, not the first acceptable one.

So don't settle for a single pass: spend the resources to explore alternatives — competing ideas, rival approaches, parallel pieces of one implementation — then converge to one coherent result. That's a synthesized answer or merged working code depending on the task; for code, "best" includes that it runs.

Breadth isn't correctness. Resolve conflicts and integrate yourself — never relay raw agent dumps or rubber-stamp a confident subagent. Verify when it sharpens the result (for code: tests, build, repro).

Modes

/full-send <task> — single-task, one wide round.
/full-send sustained — session posture. Only an explicit user signal drops it (full-send off, stop full-send, or any unambiguous stop signal in any language); mid-sustained questions or redirects don't. New session starts fresh.

Open the first response with the marker [full-send] / [full-send sustained], batch any genuine ambiguity about what is asked once — no "should I go wide?" padding — then go.

Judgment, not phases

No fixed procedure — use judgment on how wide to go and how to slice the work. Defaults: