Problem solved

People routinely give AI models vague or ambiguous instructions and then blame the model when the output misses the mark. Harfi isolates that skill by removing the ability to write the answer, forcing players to specify shape, length, format and vocabulary precisely enough that a strict parser accepts the model's output. It turns prompt specification into a measurable, repeatable exercise.

Use cases

• Practising writing unambiguous, tightly scoped instructions for LLMs
• Learning to constrain output length, bullet counts and formatting explicitly
• Building intuition for JSON, CSV and YAML output shaping without writing code
• Testing whether a prompt that works on one model also works on another
• Turning prompt-engineering drills into a scored, replayable game
• Illustrating in a classroom or workshop why literal specifications matter

Audience

• Developers
• Product Managers
• Content Creators
• Hobbyists
• Researchers

Pros

• Deterministic scoring by a parser in code rather than an LLM judge, so the same attempt always yields the same result
• Clear, granular feedback identifying exactly which rule failed and by how much
• First set of eight trials is free forever with no account or card required
• Progression system with par, birdie, ace and minimalist medals adds replay value
• Plain-English prompts mean no coding knowledge is needed to play

Cons

• Only forty trials in total, which experienced prompt engineers may exhaust quickly
• Full content requires a one-time $29 purchase; the free tier covers only one of five sets
• Limited to text-output constraints and does not cover multimodal or agentic prompting
• No API, library or exportable tooling — it is a game, not a production utility
• Model choice is fixed by the platform and may not reflect the models a user works with