Debate Kickoff (Starter Pack)

SkillAI & models

Lets your agent turn a technical decision into a structured debate between different AI models and get a recommendation.

Instructions available. Your AI can read the instructions. Execution depends on the setup they require.

Add ahel to your AI once: Claude, ChatGPT, Cursor, Claude Code or Codex. Then ask it to use this.

Then ask your AI: use the Debate Kickoff (Starter Pack) skill

About this skill

Starter: frame a decision as a multi-model debate, picks sides, seats providers, and launches /octo:debate with a well-formed motion

What this skill tells your AI

The instructions your AI receives, as published by nyldn/claude-octopus in skills/octopus-starter-pack/debate-kickoff/SKILL.md and read by ahel’s review.

Turn a loosely stated decision ("Redis or Memcached?", "monorepo vs polyrepo") into a well-formed multi-model debate.

When to use

The user has a two-or-more-sided technical decision and wants adversarial perspectives from different model families rather than one model's opinion.

Steps

  1. Extract the motion. Restate the user's question as a single debatable proposition (for example: "This project should use Redis over Memcached for session storage"). Confirm silently from context; do not interrogate the user.
  2. Check seats. Run ${CLAUDE_PLUGIN_ROOT}/scripts/helpers/check-providers.sh and note which providers are available. A debate needs at least two seatable non-Claude providers for cross-lab disagreement; if only Claude is available, say so and offer a single-model pro/con instead.
  3. Assign sides. Give each available provider an explicit stance (affirmative, negative, or skeptic). Prefer cross-lab pairings (for example Codex affirmative, Antigravity negative) so disagreement is structural, not stylistic.
  4. Launch. Invoke /octo:debate with the motion and the side assignments. Display the standard Octopus banner with provider indicators before dispatch.
  5. Synthesize. After rounds complete, summarize where the models converged, where they genuinely disagreed, and give one recommendation with the strongest surviving argument.

Guardrails

  • External seats cost money; state the expected cost band from the CLAUDE.md cost table before dispatch.
  • Never fabricate a provider's position. If a seat fails, report the failure and continue with the remaining seats.

Signals

GitHub stars
4k
Forks
390
Last commit
Sep 2026

Others that do the same job

Advanced
Item type
skill
Key
debate-kickoff-nyldn
Source
github.com/nyldn/claude-octopus