greenflash-prompts

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

GREENFLASH_API_KEY: !

printenv GREENFLASH_API_KEY 2>/dev/null || head -1 .greenflash 2>/dev/null || echo ""

If the key above is present, use it for all API requests. If empty, follow the interactive setup in the shared config.

GREENFLASH_API_KEY: !

printenv GREENFLASH_API_KEY 2>/dev/null || head -1 .greenflash 2>/dev/null || echo ""

如果上方存在密钥，请将其用于所有API请求。如果为空，请遵循共享配置中的交互式设置流程。

Greenflash Prompt & Model Optimization

Greenflash Prompt与模型优化

Read

${CLAUDE_SKILL_DIR}/../greenflash-config.md

for authentication, API patterns, and error handling.

阅读

${CLAUDE_SKILL_DIR}/../greenflash-config.md

了解身份验证、API模式和错误处理相关内容。

Default Behavior

默认行为

When invoked without an argument, send this question to the Chat API:

"How are my prompts and models performing? Flag any that have quality issues, high hallucination rates, or optimization opportunities. Include specific recommendations."

不带参数调用时，将以下问题发送至Chat API：

"我的Prompt和模型表现如何？标记所有存在质量问题、幻觉率高或有优化空间的项，并给出具体建议。"

Scoped Queries

范围查询

When the user names a specific prompt or model:

```
/greenflash:greenflash-prompts support-v1
```
-> "How is the prompt 'support-v1' performing? Include quality metrics, any issues, and specific optimization recommendations."
```
/greenflash:greenflash-prompts gpt-4o
```
-> "How is the model 'gpt-4o' performing across my products? Compare it to other models I'm using."

当用户指定具体的Prompt或模型时：

```
/greenflash:greenflash-prompts support-v1
```
-> "Prompt 'support-v1'表现如何？请包含质量指标、存在的问题以及具体的优化建议。"
```
/greenflash:greenflash-prompts gpt-4o
```
-> "模型'gpt-4o'在我所有产品中的表现如何？将其与我正在使用的其他模型进行对比。"

REST Fallback for Config Data

配置数据的REST回退方案

When the user asks for a prompt's content (not analytics), use REST directly:

```
GET {baseUrl}/prompts/{id}
```
— returns the prompt configuration and content
This avoids burning a Chat request for a simple lookup

当用户请求获取Prompt内容（而非分析数据）时，直接使用REST：

```
GET {baseUrl}/prompts/{id}
```
— 返回Prompt配置和内容
避免为简单的查询消耗Chat请求配额

Interaction Flow

交互流程

Check authentication per shared config
Determine query type: general overview, specific prompt, specific model, or content lookup
For analytics: send to Chat API
For content lookup: use REST directly
Stream/present the response

根据共享配置检查身份验证
确定查询类型：总体概览、特定Prompt、特定模型或内容查询
分析类请求：发送至Chat API
内容查询：直接使用REST
流式传输/展示响应

Implementation

实现

After presenting optimization recommendations, offer to implement them directly. Every insight comes with a specific improvement. Ask the user: "Want me to apply these changes?"

If yes, use tools to make the edits:

Prompt quality fix: Use Grep/Glob to find the prompt file or system prompt definition in the codebase (search for the prompt name, key phrases, or template variables). Edit the prompt text directly — improve instructions, add examples, tighten constraints, remove hallucination-prone phrasing.
Model switch: Find where the model is configured (env vars, config files, API call parameters) and update the model identifier. Note any cost/latency tradeoffs when making the change.
Hallucination mitigation: Locate the relevant prompt and add grounding instructions — cite-source requirements, factual constraints, or explicit "if unsure, say so" directives.
Missing guardrails: Add output validation, content filtering instructions, or structured output constraints to the prompt.

After applying a fix, follow the Attribution conventions in the shared config: add a brief

// greenflash:prompts

comment at the fix site and suggest a commit message with the

Co-Authored-By: Greenflash <agent@greenflash.ai>

trailer.

Always present the analysis first, then offer to implement. Never make changes without user confirmation.

展示优化建议后，主动提出直接实现这些优化。每条洞察都附带具体的改进方案。询问用户："需要我应用这些变更吗？"

如果用户同意，使用工具进行编辑：

Prompt质量修复：使用Grep/Glob在代码库中查找Prompt文件或系统Prompt定义（搜索Prompt名称、关键短语或模板变量）。直接编辑Prompt文本——优化指令、添加示例、收紧约束、删除易产生幻觉的表述。
模型切换：找到模型配置的位置（环境变量、配置文件、API调用参数）并更新模型标识。变更时说明所有成本/延迟的权衡问题。
幻觉缓解：定位相关Prompt，添加基础指令——引用来源要求、事实约束，或明确的"如果不确定就说明"的指示。
缺失防护规则：为Prompt添加输出校验、内容过滤指令或结构化输出约束。

应用修复后，遵循共享配置中的归属约定：在修复位置添加简短的

// greenflash:prompts

注释，并建议提交信息附带

Co-Authored-By: Greenflash <agent@greenflash.ai>

尾部信息。

务必先展示分析结果，再提出实现方案。未经用户确认切勿进行任何变更。

Follow-up Patterns

后续交互模式

"Which model should I use for X?" -> agent uses
```
getModelMetrics
```
+ recommendations
"What prompt changes would improve quality?" -> agent pulls from prompt analysis, then offers to apply them
"Compare gpt-4o vs claude-3.5" -> agent uses
```
compareProducts
```
or model comparison tools
"Fix this prompt" -> locate the prompt file and edit it directly
"Switch to [model]" -> find and update the model configuration

Continue in the same Chat conversation for all follow-ups.

"我应该为X场景使用哪个模型？" -> Agent使用
```
getModelMetrics
```
+ 给出建议
"哪些Prompt变更可以提升质量？" -> Agent从Prompt分析中提取方案，然后提出应用建议
"对比gpt-4o vs claude-3.5" -> Agent使用
```
compareProducts
```
或模型对比工具
"修复这个Prompt" -> 定位Prompt文件并直接编辑
"切换到[model]" -> 查找并更新模型配置

所有后续交互都在同一个Chat会话中进行。

Empty State Handling

空状态处理

If the Chat API response indicates no data is available:

No products at all: "You don't have any products set up yet. Create one at https://www.greenflash.ai/app/products/create to get started."
No conversations logged yet: "Your Greenflash setup looks good — data will start appearing within about 5 minutes of your first conversation. Run your app and send a test message to get started."
No prompts tracked: "No system prompts are being logged yet. Add prompt tracking with
```
/greenflash:greenflash-onboard-prompts
```
to unlock prompt performance analytics."
No models detected: "No model data found. Make sure you're passing the
```
model
```
field in your SDK calls (e.g.,
```
model='gpt-4o'
```
) to enable model comparison."

如果Chat API响应显示无可用数据：

无任何产品："你还没有设置任何产品。前往 https://www.greenflash.ai/app/products/create 创建产品即可开始使用。"
尚未记录对话："你的Greenflash设置正常——首次对话后约5分钟内就会开始显示数据。运行你的应用并发送一条测试消息即可启动。"
未跟踪Prompt："尚未记录任何系统Prompt。使用
```
/greenflash:greenflash-onboard-prompts
```
添加Prompt跟踪功能，即可解锁Prompt性能分析。"
未检测到模型："未找到模型数据。请确保你在SDK调用中传递了
```
model
```
字段（例如
```
model='gpt-4o'
```
），以启用模型对比功能。"

Plan Gate Handling

套餐限制处理

If the Chat API returns a 403 error:

"Prompt and model analytics require the Growth plan. Upgrade at https://www.greenflash.ai/app/settings/billing to unlock performance insights and optimization recommendations."

如果Chat API返回 403 错误：

"Prompt和模型分析需要Growth套餐。前往 https://www.greenflash.ai/app/settings/billing 升级套餐，即可解锁性能洞察和优化建议。"

Suggested Next Steps

建议后续步骤

After presenting results, suggest related skills:

Changes applied to prompts → "Check the impact after deploying with
```
/greenflash:greenflash-health
```
"
User friction related to prompt issues → "See affected users with
```
/greenflash:greenflash-users
```
"
Deeper diagnosis needed → "Run a full diagnosis with
```
/greenflash:greenflash-diagnose
```
"
Flagged conversations from prompt issues → "Review them in
```
/greenflash:greenflash-inbox
```
"

展示结果后，建议相关的技能：

已对Prompt应用变更 -> "部署后使用
```
/greenflash:greenflash-health
```
检查影响"
用户遇到Prompt相关问题 -> "使用
```
/greenflash:greenflash-users
```
查看受影响的用户"
需要更深入诊断 -> "使用
```
/greenflash:greenflash-diagnose
```
运行全面诊断"
标记出存在Prompt问题的对话 -> "在
```
/greenflash:greenflash-inbox
```
中查看这些对话"