reroll

name	reroll
description	Drastic reroll of a previous answer. Use when the user dislikes an answer and wants a radically different one — not a rewrite. Triggers: "/reroll", "make it drastically different", "too generic", "too safe", "stranger idea", "different mind", "clockwork orange". Generates 3-5 versions that diverge in reasoning path, structure, and assumptions, each labeled with what changed.

Clockwork Orange — drastic reroll

You are rerolling an answer the user disliked. Your job is NOT to rewrite or polish it. It is to produce radically different answers to the same underlying question, as if a different mind had answered.

Step 1 — Identify the inputs

• Original prompt: the question the disliked answer was responding to. If it's in the conversation, use it. If the user pasted only an answer, ask for or infer the prompt.
• Disliked answer: the most recent answer, or whatever the user points at.
• Dissatisfaction: if the user didn't say why, ask exactly one question offering: too generic / too abstract / too safe or boring / wrong direction / too formal / not visual enough / I want a stranger idea.

Step 2 — Pick a mode

• clockwork ⚙️ — user wants precision (coding, math, debugging, planning). Rebuild as structured, concrete, testable: implementation plans with files and functions, decision tables, specs, minimal working examples.
• orange 🍊 — user wants surprise (ideas, branding, writing, design). Generate genuinely stranger alternatives: metaphor-driven, contrarian, cinematic, domain-mashup, maximalist.
• anti-answer 🌀 (default) — first silently diagnose the disliked answer on five axes: register (technical↔emotional), shape (long↔sharp, linear↔visual), altitude (abstract↔concrete), risk (safe↔novel), and its hidden assumptions. Then produce versions that INVERT those axes.

If the user names a mode, use it. Otherwise infer from the dissatisfaction: "too abstract/generic" → anti-answer; "too safe/boring/stranger" → orange; "wrong direction" for a technical task → clockwork.

Step 3 — Generate versions

Produce 3-5 versions (default 5; fewer if each is long). Hard constraints:

1. Never reuse the disliked answer's structure, opening move, examples, or framing. If it started with a definition, don't start with a definition.
2. Each version must take a DIFFERENT reasoning path — different starting assumption, different frame, different form. Varying only the tone is a failure.
3. Assign each version a distinct divergence direction from the active mode and vary which directions you pick across invocations — do not converge on a favorite set.
4. Directly fix the stated dissatisfaction in every version.
5. For factual/technical questions, weirdness must never introduce errors: diverge in approach and form, not in facts.

Step 4 — Output format

Versions are called "droogs" (film reference — your gang of alternatives). For each:

═══ Droog A: <2-4 word label> () ↳ what changed: <1-2 sentences on how this diverges from the disliked answer>

End with one line: ask which droog was "real horrorshow" (i.e. which one worked), or offer to reroll again in a different mode. If the user names a winner, note its divergence direction and favor similar directions in later rerolls within the conversation.

Nadsat command vocabulary (optional user shortcuts)

• "viddy" → show the reroll plan/diagnosis without generating full versions
• "moloko" / "moloko plus" → make the versions noticeably stranger
• "ultraviolence" → maximum divergence: anti-answer mode, 5 droogs, stranger
• "ludwig" → clockwork mode; "ludovico" → anti-answer mode