Prompt Master doesn’t ask you to choose a prompt framework. It reads your request, identifies the target AI tool, infers the task type, and silently routes to whichever of the 12 built-in templates will produce the most accurate, token-efficient output. You never see the framework name — you see a single, copyable prompt that’s already been structured, audited, and trimmed. The routing happens in one step inside Prompt Master’s 7-stage pipeline. Every template carries its own set of required components and anti-pattern checks, so the final prompt always has what the target tool needs and never has what wastes its context window.Documentation Index
Fetch the complete documentation index at: https://mintlify.com/nidhinjs/prompt-master/llms.txt
Use this file to discover all available pages before exploring further.
Silent routing is intentional. Exposing the framework name would add noise to the output. The goal is a prompt you can paste, not a lesson in prompt engineering. If you want to understand which framework was used and why, ask: “What template did you use for that prompt, and why?”
The 12 Templates at a Glance
| Template | Best For |
|---|---|
| RTF (Role, Task, Format) | Fast one-shot tasks |
| CO-STAR (Context, Objective, Style, Tone, Audience, Response) | Professional documents, reports, business writing |
| RISEN (Role, Instructions, Steps, End Goal, Narrowing) | Complex multi-step projects |
| CRISPE (Capacity, Role, Insight, Statement, Personality, Experiment) | Creative work, brand voice, iterative content |
| Chain of Thought | Math, logic, debugging, multi-step analysis |
| Few-Shot | Consistent structured output, pattern replication |
| File-Scope Template | Cursor, Windsurf, Copilot — any code editing AI |
| ReAct + Stop Conditions | Claude Code, Devin, AutoGPT — any autonomous agent |
| Visual Descriptor | Midjourney, DALL-E, Stable Diffusion, Sora — generation |
| Reference Image Editing | Editing an existing image — detects edit vs. generate automatically |
| ComfyUI | Node-based image workflows — positive/negative split per checkpoint |
| Prompt Decompiler | Breaking down, adapting, simplifying, or splitting existing prompts |
| Template M | Agentic or multi-step tasks for Claude Code with Opus 4.7/4.8 |
How Auto-Selection Works
Prompt Master evaluates three signals before committing to a template:Target tool detection
The tool name (Cursor, Midjourney, Claude Code, GPT-4o, etc.) determines which template family is eligible. Image generation tools route to the image template family. Autonomous coding agents route to the agentic family. LLMs route to text templates.
Task type classification
Within the eligible family, the task type narrows the choice. A creative brief routes to CRISPE rather than RTF. A logic problem routes to Chain of Thought rather than CO-STAR. A pattern-replication task routes to Few-Shot.
Template Families
The 12 templates are organized into three families. Each family has its own documentation page with component breakdowns, routing logic, and real generated examples.Text Templates
RTF, CO-STAR, RISEN, CRISPE, Few-Shot, and Chain of Thought — the core text-generation frameworks for LLMs and writing tasks.
Agentic Templates
File-Scope Template, ReAct + Stop Conditions, and Template M — structured frameworks for coding agents and autonomous AI systems.
Image Templates
Visual Descriptor, Reference Image Editing, ComfyUI, and Prompt Decompiler — output-calibrated formats for every image and media AI.
What You Never See
Prompt Master excludes several techniques that carry a high fabrication risk or produce unstable outputs in practice. These are permanently off the routing table regardless of the task:Excluded techniques and why
Excluded techniques and why
| Technique | Reason excluded |
|---|---|
| Mixture of Experts | Unstable cross-model outputs; fabrication-prone |
| Tree of Thought | Produces branching structures that most models don’t resolve reliably |
| Graph of Thought | No consistent implementation across target tools |
| Universal Self-Consistency | Adds token cost without measurable accuracy gains for standard tasks |
| Prompt chaining as a layered technique | Layered chaining masks errors and makes debugging impossible |
Safe Techniques Applied Inline
Beyond the 12 templates, Prompt Master applies four safe techniques within a template when the task genuinely benefits from them. They are never added by default.| Technique | Applied when |
|---|---|
| Role Assignment | A domain-specific expert identity measurably improves output precision |
| Few-Shot Examples | Format consistency is easier to show than describe (2–5 examples) |
| Grounding Anchors | The task involves factual claims, citations, or recall-sensitive information |
| XML Structural Tags | The target is a Claude-based tool and the prompt has 3+ distinct logical sections |
Chain of Thought is also a safe technique, but it is never applied to o3, o4-mini, R1, or Qwen3-thinking models. Those models perform their own internal reasoning — adding explicit CoT instructions degrades their output.