meta-prompter

安装量: 54
排名: #13644

安装

npx skills add https://github.com/delexw/claude-code-misc --skill meta-prompter

Meta-Prompter Prompt evaluation and optimization using the meta-prompter-mcp CLI. Returns for the caller to execute. Inputs Raw arguments: $ARGUMENTS Infer from the arguments: CONTEXT: the task context or prompt to optimize OUT_DIR: output directory, or .implement-assets/meta-prompter if not provided Session If starting fresh: create SESSION_ID = sess-YYYYMMDD-HHMMSS and use it for all files. Execution See references/rules.md for CLI usage, environment variables (model configuration), and error handling. Execute all steps A through E: A) Evaluate If tag doesn't exist: Run: npx meta-prompter-mcp "CONTEXT" via Bash, with --model flag per references/rules.md model resolution Save the JSON output to and include: "original_prompt": "CONTEXT" Otherwise, skip to Clarify. B) Clarify If tag is missing or incomplete: Read the 3 questions from Attempt to answer all questions based on context If you don't have enough context to answer: Use AskUserQuestion tool to ask each question Save the questions and answers to tag as: { "answers" : [ { "q" : "" , "answer" : "" } , { "q" : "" , "answer" : "" } , { "q" : "" , "answer" : "" } ] } C) Build PROMPT Proceed only when all 3 answers are present and clear. Build PROMPT : Base on rewrite from evaluation result Append a short Clarifications section from the 3 answers (concise bullet points). D) Re-evaluate and gate Think the re-evaluation process carefully Re-run npx meta-prompter-mcp "" via Bash (with --model flag per references/rules.md model resolution) Overwrite with this latest evaluation JSON result (preserve "original_prompt" and, if present, "contextual_prompt" ). When global < 8 : STOP execution immediately. Do NOT invent questions. Use the 3 questions returned by this re-evaluation . Redo B) Clarify , redo C) Build PROMPT , and re-run this step (D) . Repeat until global >= 8 , then proceed using the . Max 3 iterations. If global < 8 after 3 attempts, use the best-scoring prompt as and note the score. E) Return If OUT_DIR is provided: Run mkdir -p OUT_DIR via Bash Use the Write tool to save to OUT_DIR/soul.md Return to the caller for execution

返回排行榜