Full Send

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 — single-task, one wide round.
• /full-send sustained — session posture, persists until explicit off-signal.

Off-signal (sustained)

Only an explicit user signal drops it: full-send off, stop full-send, or any unambiguous stop/return signal in any language. Mid-sustained questions or redirects don't drop it. New session starts fresh.

Preamble — first response after /full-send

Open with the marker: [full-send] (single) or [full-send sustained]. Surface task ambiguity in one batch only if it's genuinely unclear what is asked — 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:

• Every agent earns its spend. Name its distinct contribution — a different file, angle, approach, or risk surface — or don't spawn it. When you can't name the next one, the fan-out is done.
• Open-ended work goes in rounds. Fan out, converge, then spawn another round only if the last earned it.
• Brief every subagent. They start empty — pass the skills, project context, and prior decisions they need, or they drift onto a neighboring question.
• Workflows stay opt-in. Agents and teams are the default instrument; reach for the Workflow tool only when the user names it (e.g. /full-send workflows).

Skip the fan-out when it can't change the outcome: deterministic single step (lint, rename), atomic lookup (one fact), or strictly sequential subproblems (B needs A's output first). Say so once instead of spawning agents that produce nothing.

Stacking

• max-effort: full-send owns generation (fan out, converge); max-effort owns the adversarial pass on the converged result. Build wide, then tear down — in that order.
• afk: the subagent budget picked in the AFK preamble is the spend authority — fan out within it, never past it.

Final summary

Inline, marker-led: [full-send] / [full-send sustained]. In sustained mode the marker is load-bearing — it's how the posture survives context compression, so it appears on every turn that emits a summary. After it, open items only: one 🚩 line per angle left unexplored, agent that came back empty, or decision waiting on you. If nothing's open: ✅ nothing open.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: ikkeseb
• Source: ikkeseb/skills
• License: MIT

Install and usage instructions live in the source repository linked above.