A Claude skill that reverse-engineers prompts for 30+ AI tools

What it does

Prompt Master is a Claude skill—essentially a carefully engineered system prompt—that interviews you, picks the right prompt architecture, and outputs a tight, tool-specific prompt you can paste into Claude, ChatGPT, Cursor, Midjourney, or roughly 30 other systems. It runs a seven-step pipeline inside Claude: detects the target tool, extracts intent across nine dimensions, asks up to three clarifying questions, selects one of twelve internal templates, and finally runs a “token efficiency audit” to strip fluff.

The interesting bit

The skill treats prompt engineering as translation, not decoration. Its core thesis is that every word should be load-bearing, so it converts vague requests like “make it look like Notion” into exact hex codes, spacing units, and animation timings. For tools it has never seen, it falls back to a “Universal Fingerprint”—four diagnostic questions—rather than guessing a format.

Key highlights

• Covers 30+ tool profiles, from reasoning LLMs to video generators, each with tool-specific syntax rules (e.g., Midjourney comma-separated descriptors, Stable Diffusion weight syntax).
• Auto-selects from 12 prompt frameworks (RTF, CO-STAR, RISEN, CRISPE, Chain of Thought, Few-Shot, etc.) without exposing the machinery to the user.
• Explicitly strips chain-of-thought instructions for models like o3 and DeepSeek-R1 that think internally, avoiding wasted tokens and confused outputs.
• Outputs include metadata: estimated token weight, target tool, and a one-line strategy note explaining why the structure was chosen.

Caveats

• This is a meta-prompt packaged as a skill, not a codebase; the repository appears to contain the prompt definition and documentation rather than executable software.
• The README claims “full context and memory retention,” but it is unclear how memory persists across separate Claude conversations since skills typically start fresh each session.
• Token counts like “Light (~60)” in examples are estimates; the README does not explain how they are calculated or validated.

Verdict

Worth a look if you spend real money on API calls and suspect half your tokens are spent correcting ambiguous instructions. Skip it if you already maintain your own prompt library or prefer to hand-craft every system prompt yourself.