From harness-engineering
Improve and enhance prompts for LLM interactions. Use when user says "improve this prompt", "make this prompt better", "enhance prompt", "review my prompt", or shares a prompt and asks for feedback.
How this skill is triggered — by the user, by Claude, or both
Slash command
/harness-engineering:prompt-enhancerThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Run the PTC analysis script:
Run the PTC analysis script:
python3 ${CLAUDE_PLUGIN_ROOT}/skills/prompt-enhancer/scripts/enhance.py analyze "<prompt_text_or_file_path>"
Returns JSON with scores and suggestions:
{
"has_role": false,
"has_examples": false,
"has_output_format": true,
"specificity_score": 3,
"word_count": 42,
"suggestions": ["Add role definition", "Include few-shot examples"]
}
Apply surgical improvements based on analysis:
Do NOT over-engineer. If a dimension is already good, leave it.
Run formatting script:
python3 ${CLAUDE_PLUGIN_ROOT}/skills/prompt-enhancer/scripts/enhance.py format \
--original "<original>" --enhanced "<enhanced>"
Show to user:
Input: "Summarize this text" Output: Enhanced with role ("You are a concise technical writer"), output format ("Provide a 3-bullet summary"), and constraints ("Focus on actionable insights, not background").
Input: "You are a code reviewer. Review the following Python code for bugs and suggest fixes. Format output as a table with columns: Line, Issue, Fix." Output: Added 1 few-shot example showing expected table format. Other dimensions left unchanged.
npx claudepluginhub emingenc/harness-engineering --plugin harness-engineeringAnalyze and improve existing prompts for better performance
Optimizes weak or vague prompts into structured, precision-engineered instructions using RSCIT, chain-of-thought, and few-shot frameworks. Reduces hallucinations and token usage across any LLM.
Transforms rough prompts, task descriptions, or jobs into optimized AI instruction prompts using best practices. Activates on requests to improve, optimize, or refine prompts for Claude/GPT.