API 代理响应判定安全规范
SkillCloud & infraYour AI can judge gateway, proxy, WAF, and CDN responses without mistaking ordinary text for errors. This skill provides safe rules for response checks and keyword matching, so terms like Cloudflare, 502, bad gateway, or error that appear inside a normal long body are no longer treated as an upstream failure. It applies to middleware code such as nginx, openresty, rewrite, and redirect logic.
Available today. Use it from your connected AI after setup.
No other account needed.
After adding the skill, point your AI at the proxy, gateway, WAF, or CDN code you want it to write or check. Ask it to judge responses or match keywords and it will follow the safe rules automatically.
Then ask your AI: use the API 代理响应判定安全规范 skill
What your AI can do with it
- Judge proxy, gateway, WAF, and CDN middleware responses accurately
- Match keywords without flagging technical terms in long response bodies as upstream errors
- Write and review nginx and openresty code with safe error checks
- Apply the same safe response judgment to rewrite and redirect code
- Reduce false error alerts when content legitimately mentions Cloudflare, 502, or bad gateway
What this skill tells your AI
The instructions your AI receives, as published by doccker/cc-use-exp in .codex/skills/cc-api-proxy-safety/SKILL.md and read by ahel’s review.
网关、反向代理、WAF、CDN 中间件实现响应判定时,不要只靠正文关键词判断上游是否失败。关键词可以作为线索,不能替代 HTTP 状态、响应头、内容类型和业务响应结构。
核心原则
按以下优先级判定响应性质:
- 传输层和 HTTP 状态:连接失败、超时、非 2xx/3xx 状态码优先作为错误证据。
- 响应头和内容类型:
Content-Type、Server、Via、CF-Ray、认证/限流头等比正文关键词更可靠。 - 业务响应结构:先解析 JSON envelope、error code、message、data 等契约字段。
- 正文启发式:只有在无法按结构判断时,才使用关键词或 HTML 片段,并限制长度和上下文。
禁止:对完整业务正文做纯 strings.Contains(lowerBody, "cloudflare") / "bad gateway" / "error" 判错。
强弱特征拆分
强特征
强特征应同时具备明确上下文,例如:
- HTTP 状态为 502/503/504/429,且正文是上游错误页或错误 envelope。
Content-Type: text/html,正文包含错误页标题,且响应头或状态也指向 CDN/WAF/网关。- 结构化 JSON 明确给出上游错误字段,例如
{"error": {"type": "upstream_error"}}。
不要把单个短片段直接当强特征。<html、cf-ray:、cloudflare、502 bad gateway 都可能出现在技术文章、日志展示或用户输入中。
弱特征
以下内容只能作为弱证据:
- CDN/网关品牌词:
cloudflare、openresty、nginx - 短错误短语:
bad gateway、502 bad gateway、upstream error - 通用错误词:
error、reconnecting、too many requests
弱特征只有在满足以下条件时才可参与判错:
- 响应体很短,例如
<= 512字节。 - 响应结构不是合法业务 payload。
- 或者它和错误 HTTP 状态、错误响应头、错误 content-type 同时出现。
判定模板
func isUpstreamFailure(resp *http.Response, body []byte) bool {
if resp == nil {
return true
}
status := resp.StatusCode
contentType := strings.ToLower(resp.Header.Get("Content-Type"))
if status >= 500 || status == http.StatusTooManyRequests {
low := strings.ToLower(string(body))
return looksLikeGatewayError(contentType, low, len(body))
}
if isValidBusinessPayload(contentType, body) {
return false
}
if len(body) > 512 {
return false
}
low := strings.ToLower(string(body))
if len(body) <= 512 && containsWeakGatewayKeyword(low) {
return true
}
return false
}
实现时把 looksLikeGatewayError、isValidBusinessPayload、containsWeakGatewayKeyword 拆开测试,不要把所有条件写成一个大 if。
场景策略
| 场景 | 策略 |
|---|---|
| HTTP 非 2xx/3xx | 优先按状态码和响应头判错,正文用于补充分类 |
| 业务 JSON 响应 | 先解析契约字段,不要因为 data 或正文里有技术词就判错 |
| 长文本/文章/模型输出 | 不允许仅凭弱关键词判错 |
| 短错误消息 | 可使用弱关键词,但仍应结合调用上下文 |
| 流式响应 | 等待完整事件、完整 JSON chunk 或明确错误事件后再判定 |
回归测试
至少覆盖:
- 长正文包含
Cloudflare/502 Bad Gateway/bad gateway技术术语时不误判。 - 短响应体只有
502 Bad Gateway且没有业务结构时判为错误。 status=200且合法业务 JSON 中出现弱关键词时不误判。status=502且 HTML 错误页时判为错误。- streaming 半截 chunk 命中关键词时不立即判错。
与相关 skill 的边界
| skill | 关系 |
|---|---|
cc-external-system-debugging | 上游方法论:遇到 CDN/WAF/代理异常时先抓真实 request/response |
cc-api-design-safety | 本 skill 关注代理判定逻辑,不替代 API envelope 设计规范 |
cc-ops-safety | 本 skill 不处理运维审批、危险命令或生产配置变更 |
规则溯源
> 📋 本回复遵循:`cc-api-proxy-safety` - [章节]
Signals
- GitHub stars
- 1k
- Forks
- 112
- Last commit
- Aug 2026
Advanced
- Catalog kind
- skill
- Gateway key
cc-api-proxy-safety- Source
- github.com/doccker/cc-use-exp