Diagnoses and refines prompts by improving structure, techniques, and wording.
npx skills add https://github.com/psw7205/skills --skill prompt-psw7205
Please help me install this Agent Skill. Skill: prompt Source: https://github.com/psw7205/skills/tree/main/skills/prompt Command: npx skills add https://github.com/psw7205/skills --skill prompt-psw7205
High-intent search queries and answers about installing and using this skill.
Diagnose the prompt across three axes: structure (missing Instruction, Context, Input Data, or Output Indicator), technique fit, and wording clarity. Fix only the weak areas, then verify against a checklist covering delimiters, output format, and fallback responses.
Start with the simplest approach and escalate only when needed: zero-shot for clear tasks, few-shot for classification with format consistency, CoT for multi-step reasoning, self-consistency for higher reliability, and prompt chaining for multi-stage workflows.
No. Reasoning models perform their own internal reasoning, so manual CoT phrases like "step by step" interfere and should be removed. Keep instructions simple and direct, state high-level constraints, and prefer XML-structured output.
Avoid extra techniques when zero-shot instructions already produce adequate results. Adding CoT to simple classification tasks makes responses slower and less stable, and few-shot examples with imbalanced label distributions bias the output.
Separate user input from system instructions structurally by wrapping user data in XML tags like <user_query>. Place the instruction at the start, use delimiters between sections, and keep critical constraints at the beginning and end of the prompt.
ed3daiCreate specialized subagents for Claude Code plugins with a standard template.
nimadorostkarDesign, test, and refine prompts for large language models.
coval-aiAnalyze human versus LLM judge annotation discrepancies to refine evaluation prompts.
Factory-AIRefine Claude prompts into structured formats for improved reliability.
