---
name: lateral
description: "Inject random creative perspectives into any question by launching parallel agents, each given a random dictionary word as a lens. Use when a line, angle or diagnosis feels stale or obvious, or when /research or /strategy would benefit from creative reframing. Triggers on 'lateral', 'random angles', 'fresh perspectives', 'creative reframe', 'break it open'."
argument-hint: "[count] [topic or question to explore]"
allowed-tools: Bash(python3 *)
---

# Lateral Thinking via Random Injection

Launch N parallel agents, each given a random dictionary word as a creative lens on a question. The randomness forces connections that linear thinking misses.

## Process

1. **Sample words.** Default 10 agents; `/lateral 5 <topic>` changes the count.

   ```bash
   python3 ~/.claude/skills/lateral/scripts/sample-words.py 10
   ```

   Obscure words are fine. Obscurity forces more creative interpretation. Do not re-roll for "better" words. If a word is obscure, add a one-line definition to that agent's prompt.

2. **Launch all agents in one message**, one Agent call per word, `model: sonnet`. Prompt template:

   ```
   You are contributing one perspective to a multi-agent creative synthesis.

   **Topic**: {the user's question or topic}

   **Your assigned random word is: {WORD}**

   Use this word as a creative lens, metaphor, or unexpected connection point to generate a fresh insight about the topic. Interpret the word however produces the most interesting thought: literally, metaphorically, historically, scientifically, etymologically.

   Rules:
   - Respond in exactly 2-3 tight paragraphs
   - Take a clear position; do not hedge
   - Be original; the value is in unexpected connections
   - Do not explain the exercise or comment on the process
   ```

3. **Synthesise.** Name the 3-5 strongest threads, note where two random words converged on the same insight from different angles, and state the consensus or the key tension. Chris wants the distilled insight, not ten essays.

## Who calls this

- `/research` after vault search, when the answers are predictable.
- `/strategy` on the diagnosis or guiding policy before the stress test.
- Agents, when a line still feels stale after the spark-table method in `~/vault/agents/references/voice-spark-tables.md`.

#!/usr/bin/env python3
"""Sample random words from the dictionary for /lateral skill."""

import argparse
import os
import random
import subprocess
import sys

SKILL_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
DICT_PATH = os.path.join(SKILL_DIR, "words.txt")
DICT_URL = "https://raw.githubusercontent.com/dwyl/english-words/master/words_alpha.txt"


def ensure_dict():
    if os.path.exists(DICT_PATH):
        return
    print(f"Downloading dictionary to {DICT_PATH}...", file=sys.stderr)
    subprocess.run(["curl", "-s", "-o", DICT_PATH, DICT_URL], check=True)


def sample(n=10, min_len=4, max_len=9):
    ensure_dict()
    words = [
        w.strip()
        for w in open(DICT_PATH)
        if w.strip().isalpha() and min_len <= len(w.strip()) <= max_len
    ]
    return random.sample(words, min(n, len(words)))


if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument("n", type=int, nargs="?", default=10, help="Number of words")
    parser.add_argument("--min-len", type=int, default=4)
    parser.add_argument("--max-len", type=int, default=9)
    args = parser.parse_args()

    for word in sample(args.n, args.min_len, args.max_len):
        print(word)