hle-reasoning-wrapper
от wanng-ide
Health & FitnessКратко
Wraps HLE benchmark questions in a structured Chain-of-Thought (CoT) reasoning process.
Описание
Что делает этот навык
Wraps Humanity's Last Exam (HLE) benchmark questions in a Chain-of-Thought reasoning format. It injects a required Thought/Answer structure into prompts and validates that LLM responses comply with that format.
Handles both prompt formatting and output validation in one module, removing the need to manually enforce CoT structure on each HLE question.
Когда его использовать
- Evaluating LLM performance on HLE benchmark questions
- Preprocessing HLE prompts before sending to a language model API
- Validating that model outputs follow required CoT structure
- Running automated HLE benchmark tests with format compliance checks
- Debugging reasoning failures in structured LLM evaluations
Пример сценария
User asks: Answer this HLE question: What is the speed of light?
- 1
Format the question using hle.formatPrompt() to inject the required Thought/Answer structure
- 2
Send the formatted prompt to an LLM API
- 3
Receive the raw LLM response
- 4
Validate the response using hle.validateOutput() to check format compliance
Validated response confirming the model followed the Thought/Answer reasoning structure
Установка
clawhub install wanng-ide/hle-reasoning-wrapperВыполните команду CLI или вставьте промпт в Claude Code, Cursor, Codex — любой кодинг-агент, которым пользуетесь.