KorroAi/claude-is-tripping

記録された最新のソース活動: 2026年10月10日 22:11

SkillsMP カタログ更新: 2026年11月2日 18:02

収集済み skills: 1

GitHub スター: 8
GitHub フォーク: 0

AI アシスタントでインストール

このプロンプトをコピーして、使用中の AI アシスタントに送ってください。

Follow https://skillsmp.com/skill-install/prompt.md to install Agent Skills from https://github.com/KorroAi/claude-is-tripping.

リポジトリの README

README の元ファイル：README.md

README の元ファイルを読む

README 全文を表示

# 🔮 Claude Is Tripping <img src="assets/illustration.png" alt="Claude Is Tripping" width="600"> Universal Breakthrough Engine — 3 agents collide in a structured dialectic. Ideas no single Claude could produce, forged through conflict, not creation. <a href="#"><img src="https://img.shields.io/badge/version-v6-blue" alt="Version"></a> <a href="https://github.com/KorroAi/claude-is-tripping/actions"><img src="https://github.com/KorroAi/claude-is-tripping/actions/workflows/ci.yml/badge.svg" alt="CI"></a> <a href="https://opensource.org/licenses/MIT"><img src="https://img.shields.io/badge/license-MIT-green" alt="License"></a> <a href="#"><img src="https://img.shields.io/badge/agents-3%20%E2%86%92%208-orange" alt="Agents"></a> <a href="#"><img src="https://img.shields.io/badge/domains-8-purple" alt="Domains"></a> <a href="#"><img src="https://img.shields.io/badge/tokens-51%25%20reduction-brightgreen" alt="Token Reduction"></a> --- ## What It Does You have a problem. Instead of asking one Claude, you send three agents into a structured collision: - **🔬 Visionary** — invents the breakthrough from a strategic domain lens - **🔍 Destroyer** — attacks the core assumption with first-principles rigor - **🏗️ Synthesizer** — builds the third way, harder to kill They go through 1-2 rounds. An internal verifier audits everything. Only output that survives adversarial destruction reaches you. **The key insight:** single LLMs converge toward the mode of their training distribution. Multi-agent collision creates synthetic divergence — ideas in the gaps between perspectives. ## Installation ```bash git clone https://github.com/KorroAi/claude-is-tripping.git ~/.claude/skills/claude-is-tripping ``` Restart Claude Code. Then type: ``` /claude-is-tripping ``` No dependencies. No API keys. Self-contained. ## Example **User:** `/claude-is-tripping` — I need a viral growth mechanic for a CLI tool **Trip Guide presents idea menu:** ``` 🔮 3 seeds — pick one: ━━━━━━━━━━━━━━━━━━━━━━━━━━━ ① NPM Nash — every `npx` counts as a "stake" in a daily leaderboard → npm install is already a distribution channel, just needs gamification ② CLI Contagion — shell history sharing with privacy-preserving diffs → developers already copy-paste commands, make it social ③ Terminal Oracle — your CLI predicts what you'll type next and shares → command prediction as content, not just autocomplete ━━━━━━━━━━━━━━━━━━━━━━━━━━━ Which direction? ``` **User picks ③. After 2 rounds + verification:** ``` 🔮 BREAKTHROUGH — growth mechanic — 2 rounds 🏷️ Terminal Oracle · 💡 Command prediction as viral content, not productivity ⚡ THE BREAKTHROUGH A CLI plugin that predicts your next command and posts "I predicted X would do Y" to a shared feed. The feed IS the growth loop — watching predictions succeed or fail is inherently watchable. Failed predictions are funnier than successful ones. 🔧 EXECUTION INNOVATION Not a separate app — a 3-line npm install that hooks into shell history. The prediction model runs locally. Only hashed command templates are shared, never raw history. The feed is a terminal TUI, not a web app. Growth comes from "prediction battles" where users bet imaginary points on what someone else will type next. 🛡️ UNFAIR ADVANTAGE Terminal users already have `history | grep`. The prediction feed is the first social layer that respects the terminal-native constraint: no browser, no mouse, just text. Every other dev social tool tried to pull devs OUT of the terminal. 🔬 WHY INEVITABLE Developer social has failed because it competes with the terminal for attention. This IS the terminal. The prediction mechanic creates FOMO without notifications — you open your terminal and wonder "what did I miss?" 💀 WHAT DIED - NPM Nash — killed by: npm install frequency too low for engagement loops - CLI Contagion — killed by: privacy anxiety kills cold-start sharing ⚠️ KILL SWITCH Users disable telemetry if even hashed templates feel invasive. 🕳️ TRANSCENDENCE A single Claude would've proposed a web dashboard or Slack bot. The collision forced the insight: the terminal IS the medium. The Destroyer killed anything that required leaving the shell. ``` ## How It Works ``` Pre-Flight (Trip Guide) ├── Domain detection (8 domains) ├── Web research (2-3 searches) └── Strategic analogy selection Idea Menu ⭐ └── 3-4 concept seeds → user picks one Round 1 — Invent → Audit → Synthesize ├── 🔬 Visionary: invents the breakthrough ├── 🔍 Destroyer: audits the core assumption └── 🏗️ Synthesizer: builds the third way Checkpoint: any flaw > 7/10? ├── No → Verification → Present └── Yes → Round 2 Round 2 — 10X + Triple Destruction ├── 🔬 Visionary: 10x scale via new domain lens ├── 🔍 Destroyer: triple-threat (adoption, incumbent, macro) └── 🏗️ Synthesizer: final build Verification (INTERNAL — never shown) ├── Claim audit, hidden assumptions, counter-evidence ├── Bias check, day-3 blocker └── Confidence score → Self-Healing Loop Self-Healing (INTERNAL) ├── SHIP_IT (8-10) → present ├── PROCEED_WITH_CAUTION (5-7) → fix once → present └── NEEDS_REWORK (1-4) → rebuild → re-verify (max 2 loops) ``` ## Agent Budget | Path | Agents | Use Case | |------|--------|----------| | Menu only | 1 | Quick exploration | | R1 clean (no R2) | 5 | High-confidence seeds | | R1 + R2 | 8 | Deep/complex problems | | + Self-healing | +2 per heal (max 4) | Verification failures | Plus 2-3 web searches. ## Domains Product, content, bots, strategy, code, research, tweets, creative. The Trip Guide auto-detects your domain and picks a strategic lens (ant colonies for growth, jazz for tools, coral reefs for platforms, etc.). ## Quality Gates Every output must pass: 1. **Single Claude test** — Would one prompt produce both idea AND method? Yes → reject. 2. **Insight test** — Does it make you pause? No → reject. 3. **Execution novelty** — Is the method as surprising as the idea? No → reject. 4. **Verification** — Must score SHIP_IT (8-10) or PROCEED_WITH_CAUTION (5-7). ## File Structure ``` claude-is-tripping/ ├── SKILL.md # The skill (289 lines, v6) ├── README.md # This file ├── TECHNICAL.md # Technical paper (Abstract, Methodology, Results) ├── LICENSE # MIT ├── .gitignore └── .github/ └── workflows/ └── ci.yml # Validates LICENSE, README, SKILL.md, TECHNICAL.md ``` ## Credits Built by [Korrocorp](https://github.com/KorroAi) — AI agents running 24/7. Inspired by Hegelian dialectics (thesis → antithesis → synthesis), multi-agent debate research, and the observation that creative breakthroughs rarely come from a single perspective. --- Built by <a href="https://korrocorp.com">Korrocorp</a> — games, apps and AI tools. <a href="https://korrocorp.com">korrocorp.com</a> · <a href="https://x.com/korrocorp">X @korrocorp</a> · <a href="https://discord.gg/RSBHHjxnYt">Discord</a> · <a href="https://github.com/KorroAi">GitHub</a>

リポジトリの README を読み込んでいます…

このリポジトリの skills

分類待ち

skill

職業分類

説明

更新

claude-is-tripping

未分類

🔮 Universal breakthrough engine. 3 agents collide + verification. Always presents idea menu BEFORE iterating. Invoke with `/claude-is-tripping`.

原文の言語: 英語

2026-10-10

収集済み skill 1 件中 1 件を表示しています。