nuke-think
Nuke Think
This skill cannot add capability (map: references/family-map.md — the discipline judgment roles across the family load). It turns every available thinking dial to maximum and enforces a protocol that recovers a large part of the mid-tier gap on decomposable judgment tasks; it does not close the gap on pure-insight tasks.
Two ways in: other agents load it as a discipline when they hold a judgment role (skeptic, spec-architect, completeness checker, reconciler), and it is invocable directly with a hard question or decision as the argument. Either way: Part 1 first, then Part 2, always both. Escalate to Part 3 only on its triggers.
Part 1 — Thinking dials (mechanical, always first)
| Platform | Dial |
|---|---|
| Claude Code, main loop | Include the word "ultrathink" in the working prompt |
| Claude Code, subagents | Pass the highest available effort (and the tier-appropriate model) on the Agent / Workflow agent() call |
| Claude API | Extended thinking with a large token budget |
| Codex / GPT CLIs | Highest reasoning-effort flag |
| No dials available | Part 2 + Part 3 carry the load alone |
When a dial exists, turn it AND run the discipline — never one without the other. A maxed dial without the protocol reproduces the same failure modes at higher cost; the protocol without the dial leaves free depth unused.