Agent Skills
› stagewise-io/stagewise
› prompt-optimization
prompt-optimization
GitHub指导构建、审查和优化LLM系统提示词,涵盖从零创建、现有提示词审计及Token压缩优化。遵循清晰性、安全性与结构化原则,提供标准架构与工作流参考。
Trigger Scenarios
编写或优化系统提示词
减少提示词Token用量
审查提示词安全性与偏见
Install
npx skills add stagewise-io/stagewise --skill prompt-optimization -g -y
SKILL.md
Frontmatter
{
"name": "prompt-optimization",
"description": "Guides building, reviewing, and optimizing system prompts and prompt templates for LLMs. Use when creating system prompts, writing prompt templates, optimizing prompt structure, reducing prompt token usage, compressing prompts, improving prompt clarity, reviewing prompts for safety and bias, or making prompts more token-efficient. Also use when the user says \"write a prompt,\" \"optimize this prompt,\" \"system prompt,\" \"prompt engineering,\" \"make this prompt better,\" \"reduce tokens,\" or \"compress this prompt.\""
}
Prompt Optimization
Build effective, safe, and token-efficient system prompts for LLMs.
How to Use This Skill
Pick the workflow that matches the task, then follow it end-to-end:
| Task | Workflow | Reference |
|---|---|---|
| Write a new prompt from scratch | Create workflow | references/workflows.md → Workflow 1 |
| Review/audit an existing prompt | Review workflow | references/workflows.md → Workflow 2 |
| Reduce token count of a working prompt | Optimize workflow | references/workflows.md → Workflow 3 |
Each workflow references the specific detail pages below as needed. Do not read all references upfront — load them when the workflow step calls for them.
Core Principles
- Clarity over cleverness — Write prompts a new employee would follow without confusion
- Positive framing — Say what to do, not what to avoid
- Structure with delimiters — XML tags or consistent separators between content types
- Progressive specificity — Role → constraints → instructions → context → format → examples → input
- Token economy — Every token must earn its place; ≥1 constraint per 5–10 tokens
- Safety by default — Prevent injection, data leakage, and bias structurally
Prompt Architecture (ordering)
1. Identity / Role — Who the model is (1–2 sentences)
2. Core behavior rules — Non-negotiable constraints (safety, tone)
3. Task instructions — What to do and how
4. Context / Knowledge — Domain data, schemas, reference docs
5. Output format — Structure, length, style requirements
6. Examples (few-shot) — 2–3 input/output pairs in <example> tags
7. Dynamic input — User content (always last, always delimited)
Key Rules
- Delimit user input — Wrap in tags like
<user_input>; never interpolate raw - Show, don't just tell — Few-shot examples beat format descriptions
- Match prompt style to output style — Markdown in prompt → markdown out; prose → prose
- Put long documents first — Reference data above instructions (30%+ quality gain)
- Quantify, don't qualify — "≤100 words" not "keep it short"
- Imperative voice — "Respond in JSON" not "You should format your response as JSON"
- Test adversarially — Injection, edge cases, ambiguous inputs before shipping
Reference Index
Load these when a workflow step calls for them:
| Reference | Contents |
|---|---|
references/workflows.md |
End-to-end workflows for create, review, and optimize tasks |
references/clarity-and-structure.md |
XML tags, delimiters, role prompting, long-context patterns |
references/output-control.md |
Format steering, verbosity control, few-shot examples |
references/safety-and-security.md |
Injection prevention, bias mitigation, data leakage, red-teaming |
references/compression-techniques.md |
High-density writing, token pruning, compression methods, density test |
references/agentic-prompts.md |
Tool use, thinking/reasoning, state tracking, subagent orchestration |
references/anti-patterns-and-checklist.md |
8 anti-patterns, design checklist, evaluation metrics |
Version History
- 45a1544 Current 2026-07-24 21:12


