Optimize AI security guard plugin #1473
Conversation
rinfx
commented
Nov 5, 2024
- 代码重构：1）gjson解析json string改为json.unmarshal方式；2）denyMessage优先级统一调整为用户自定义 > 阿里云内容安全返回建议 > 兜底内容；3）denyMessage统一进行一次序列化，防止response body错误
- 增加拦截等级设置，与内容安全对齐（high，medium，low），用户可以设置拦截阈值，默认只拦截high，同时风险分与风险等级可以在内容安全控制台进行设置
- 增加STS Token接入方式，开源用户仍需设置AK/SK，商业化用户可基于角色扮演进行授权而不是设置明文AK/SK，避免AK/SK泄漏风险
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## main #1473 +/- ##
==========================================
+ Coverage 35.91% 43.53% +7.62%
==========================================
Files 69 76 +7
Lines 11576 12320 +744
==========================================
+ Hits 4157 5364 +1207
+ Misses 7104 6621 -483
- Partials 315 335 +20 |
| @@ -67,6 +99,7 @@ type AISecurityConfig struct { | |||
| denyCode int64 | |||
| denyMessage string | |||
| protocolOriginal bool | |||
| riskLevelBar string | |||
| OpenAIStreamResponseFormat = OpenAIStreamResponseChunk + "\n\n" + OpenAIStreamResponseEnd + "\n\n" + `data: [DONE]` | ||
|
|
||
| TracingPrefix = "trace_span_tag." | ||
| // TracingPrefix = "trace_span_tag." |
| "ServiceParameters": fmt.Sprintf(`{"content": "%s"}`, marshalStr(content, log)), | ||
| } | ||
| if config.token != "" { | ||
| params["SecurityToken"] = config.token |
There was a problem hiding this comment.
在官网文档里没有找到这个参数诶，能提供一下出处吗？
另外 config.token 这个字段名字是不是用 securityToken 好一点？
https://help.aliyun.com/document_detail/2671445.html
https://help.aliyun.com/document_detail/108840.html
There was a problem hiding this comment.
SecurityToken 这个不是内容安全官方文档提供的，属于是角色扮演才会用到的，内容安全那边对外没有提供api文档
There was a problem hiding this comment.
直接在插件里面填用户的AK/SK有风险，会直接暴露在控制台上，角色扮演可以避免用户在控制台填写AK/SK
There was a problem hiding this comment.
| @@ -114,6 +113,8 @@ func (config *AISecurityConfig) incrementCounter(metricName string, inc uint64) | |||
|
|
|||
| func riskLevelToInt(riskLevel string) int { | |||
| switch riskLevel { | |||
| case MaxRisk: | |||
There was a problem hiding this comment.
嗯，因为有时候内容安全误拦截比较多，设置为max的话，就只记录检测结果，但是不会拦截请求，这样用户可以先设置为max观察一段时间，之后再调整等级