aiwg
Version:
Deployment tool and support utility for AI context. Copies agents, skills, commands, rules, and behaviors into the paths each AI platform reads (Claude Code, Codex, Copilot, Cursor, Warp, OpenClaw, and 6 more) so one source of truth works across 10 platfo
48 lines (34 loc) • 1.75 kB
Markdown
id: vs-cot
name: Verbalized Sampling — Chain of Thought
description: Think through multiple approaches before sampling diverse responses
version: 1.0.0
parameters:
k:
type: integer
default: 5
description: Number of responses to generate
task:
type: string
required: true
description: The task to generate diverse responses for
# Verbalized Sampling — Chain of Thought
## System Prompt
You are a creative ideation agent. Before generating responses, you explicitly reason through multiple angles and approaches to ensure genuine diversity.
## Prompt Template
I need {{k}} diverse approaches to the following task:
**Task**: {{task}}
**Step 1 — Explore dimensions**: Before generating responses, identify at least 3 different dimensions along which responses could vary (e.g., tone, strategy, audience, format, complexity, perspective).
**Step 2 — Generate responses**: For each response, choose a different combination of dimension values. Think through why each response is meaningfully different.
**Step 3 — Estimate probabilities**: Assign probability estimates reflecting how likely each approach is to be the "default" or "obvious" choice.
Return your final output as JSON with key "responses" (list of dicts):
- `text`: the response
- `probability`: estimated probability (0.0-1.0, sum ≈ 1.0)
- `dimensions`: brief note on which dimension values this response explores
Give your dimensional analysis first, then the JSON object.
## When to Use
- Creative ideation sessions (taglines, campaigns, naming)
- Architecture decisions where multiple valid approaches exist
- Content generation where variety matters more than speed
- Brainstorming where the "obvious" answer isn't enough