This guide helps you write clear prompts for an AI helper called Pi. Pi runs different models like DeepSeek. Good prompts give reliable results for coding, review, diagnosis, and research tasks.
The rules are simple. Use one clear task per prompt. Tell the AI what done looks like. Use numbered checklists and stable tags like <task> and <output_contract>. Keep your instructions short and concrete.
Anyone who works with AI for coding or investigation can benefit. This skill turns vague requests into structured, actionable prompts that get the job done right.
Global
mkdir -p ~/.claude/skills/pi-promptingProject
mkdir -p .claude/skills/pi-promptingSource Repository
Cavemanjuliusbrussee/caveman
Talk like a smart caveman to save tokens without losing technical accuracy
Cavemanmattpocock/skills
Caveman mode for AI cuts token count 75% without losing technical accuracy
Enhance Promptgoogle-labs-code/stitch-skills
Turn vague UI ideas into clear, polished prompts with design system consistency
Humanizer Zhop7418/humanizer-zh
Remove AI writing patterns to make your text sound natural and human
Gws Modelarmorgoogleworkspace/cli
Filter harmful user content automatically with Google Model Armor for safety
Prompt Engineering Patternswshobson/agents
Master advanced prompt engineering patterns to get reliable and powerful AI outputs every time
Context Engineeringaddyosmani/agent-skills
Optimize your agent's context to improve output quality and focus
Writetw93/waza
Rewrite your prose to sound human and natural every time