Is release-gradual safe to install?

SaferSkills independently audited release-gradual (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.

CategoryWeightCategory scoreContribution

Securityprompt, exec, net, exfil, eval

35%

100

35.0 pts

Supply chainhash, typosquat, maintainer, lockfile

20%

100

20.0 pts

Maintenancestaleness, pinning, CI

15%

100

15.0 pts

TransparencySKILL.md, perms, README

15%

100

15.0 pts

Communityinstalls, verify, response

15%

100

15.0 pts

Findings & checks · 0 flagged

Securityscore 100 · 0 findings

✓—All security checks passedNo findings in this category for the latest scan.pass

Supply chainscore 100 · 0 findings

✓—All supply chain checks passedNo findings in this category for the latest scan.pass

Maintenancescore 100 · 0 findings

✓—All maintenance checks passedNo findings in this category for the latest scan.pass

Transparencyscore 100 · 0 findings

✓—All transparency checks passedNo findings in this category for the latest scan.pass

Communityscore 100 · 0 findings

✓—All community checks passedNo findings in this category for the latest scan.pass

The primary manifest — the file an agent reads to learn what this artifact does.

pm-07-monitoring/skills/release-gradual/SKILL.md · 1 file

pm-07-monitoring/skills/release-gradual/SKILL.md20.9 KB · Markdown

name: release-gradual

description: 当需要执行灰度发布时使用。灰度发布自动执行，执行渐进式灰度发布，从1%到10%到50%到100%，各阶段自动监控指标并判断是否进入下一阶段，支持P0指标恶化时自动回滚。🤖 AI自动执行。关键词：灰度发布、渐进式发布、Feature Flag、自动回滚、发布策略、金丝雀发布、小流量、逐步放量。

metadata:

module: "产品监控与迭代"

sub-module: "发布上线"

type: "pipeline"

version: "2.0"

domain_tags: ["互联网", "通用"]

trigger_examples:

- "做灰度发布，先放1%流量"

- "帮我做小流量验证"

- "逐步放量到全量"

interaction_mode: "ai_auto"

灰度发布自动执行

核心原则

触发器驱动：由质量门禁通过事件自动触发灰度发布，指标恶化自动触发回滚
自动化验收：各阶段指标自动监控，阶段转换自动判定，回滚自动执行
持续部署：灰度策略配合Feature Flag，实现渐进式部署和即时回滚
实时复盘：各阶段指标实时汇总，发布完成后即时生成复盘输入

交互模式

🤖 AI自动执行

触发条件：

质量门禁通过
手动触发（发布负责人确认）

输入

输入项	类型	必填	来源	说明
发布内容	JSON	是	发布管理系统	待发布的版本和变更内容
Feature Flag配置	JSON	是	Feature Flag系统	灰度开关配置
监控指标定义	JSON	是	监控系统	各阶段监控指标
灰度策略配置	JSON	是	发布策略库	各阶段流量比例和持续时间

发布内容结构示例

{
  "release_id": "release_2024_0125_001",
  "version": "v2.1.0",
  "build_ref": "abc123def",
  "change_summary": "新增微信登录功能，优化登录流程",
  "affected_services": ["auth-service", "user-service"],
  "rollback_version": "v2.0.9",
  "rollback_strategy": "feature_flag_offline"
}

灰度阶段配置

标准灰度阶段

阶段	流量比例	持续时间	最小持续时间	通过条件
Stage 0	0% (验证)	-	10分钟	构建验证通过
Stage 1	1%	30分钟	30分钟	核心指标无异常
Stage 2	10%	2小时	1小时	P0指标稳定
Stage 3	50%	4小时	2小时	无新增异常
Stage 4	100%	-	-	所有条件通过

护栏指标配置

{
  "guardrail_metrics": {
    "p0_metrics": [
      {
        "name": "error_rate",
        "description": "5xx错误率",
        "unit": "percentage",
        "warning_threshold": 0.01,
        "rollback_threshold": 0.03,
        "comparison": "vs_baseline"
      },
      {
        "name": "latency_p99",
        "description": "P99响应时间",
        "unit": "milliseconds",
        "warning_threshold": 500,
        "rollback_threshold": 1000,
        "comparison": "vs_baseline"
      },
      {
        "name": "availability",
        "description": "服务可用性",
        "unit": "percentage",
        "warning_threshold": 99.9,
        "rollback_threshold": 99.0,
        "comparison": "absolute"
      }
    ],
    "p1_metrics": [
      {
        "name": "api_success_rate",
        "description": "API成功率",
        "unit": "percentage",
        "warning_threshold": 99,
        "rollback_threshold": 97
      },
      {
        "name": "error_count",
        "description": "错误次数",
        "unit": "count",
        "warning_threshold": 100,
        "rollback_threshold": 500
      }
    ],
    "p2_metrics": [
      {
        "name": "conversion_rate",
        "description": "业务转化率",
        "unit": "percentage",
        "warning_threshold": -10,
        "rollback_threshold": -30,
        "comparison": "vs_previous_phase"
      }
    ]
  }
}

执行步骤

Step 1: 灰度计划生成

#### 1.1 发布前验证

验证清单：

验证项	检查内容	失败处理
构建产物	验证构建产物完整性	阻止发布
回滚版本	确认回滚版本可用	阻止发布
Feature Flag	验证Flag配置正确	阻止发布
监控告警	验证告警规则已配置	阻止发布
值班人员	确认值班人员在线	警告

验证输出：

{
  "pre_release_verification": {
    "build_artifact": {"verified": true, "sha256": "abc123..."},
    "rollback_version": {"available": true, "version": "v2.0.9"},
    "feature_flag": {"configured": true, "flag_key": "feature_wechat_login"},
    "monitoring": {"alerts_configured": true, "channel": "pagerduty"},
    "on_call": {"confirmed": true, "engineer": "dev_zhang"},
    "ready_to_release": true
  }
}

#### 1.2 灰度计划生成

计划结构：

{
  "canary_plan": {
    "plan_id": "canary_2024_0125_001",
    "release_id": "release_2024_0125_001",
    "phases": [
      {
        "phase_id": "phase_1",
        "traffic_percentage": 1,
        "duration_minutes": 30,
        "min_duration_minutes": 30,
        "target_users": {
          "strategy": "random",
          "percentage": 1
        },
        "success_criteria": {
          "p0_metrics": "all_normal",
          "p1_metrics": "all_normal",
          "manual_approval": false
        }
      },
      {
        "phase_id": "phase_2",
        "traffic_percentage": 10,
        "duration_minutes": 120,
        "min_duration_minutes": 60,
        "target_users": {
          "strategy": "random",
          "percentage": 10
        },
        "success_criteria": {
          "p0_metrics": "stable",
          "p1_metrics": "all_normal",
          "manual_approval": true
        }
      },
      {
        "phase_id": "phase_3",
        "traffic_percentage": 50,
        "duration_minutes": 240,
        "min_duration_minutes": 120,
        "target_users": {
          "strategy": "random",
          "percentage": 50
        },
        "success_criteria": {
          "p0_metrics": "stable",
          "p1_metrics": "stable",
          "manual_approval": true
        }
      },
      {
        "phase_id": "phase_4",
        "traffic_percentage": 100,
        "duration_minutes": null,
        "target_users": {
          "strategy": "all"
        },
        "success_criteria": {
          "p0_metrics": "all_normal",
          "p1_metrics": "all_normal",
          "manual_approval": true
        }
      }
    ]
  }
}

Step 2: Feature Flag自动配置

#### 2.1 Flag配置生成

配置规则：

{
  "feature_flag_config": {
    "flag_key": "feature_wechat_login",
    "flag_type": "boolean",
    "default_value": false,
    "serving_config": {
      "phase_1": {
        "percentage": 1,
        "rules": [
          {
            "name": "random_1_percent",
            "percentage": 1,
            "variation": true
          }
        ]
      },
      "phase_2": {
        "percentage": 10,
        "rules": [
          {
            "name": "random_10_percent",
            "percentage": 10,
            "variation": true
          }
        ]
      },
      "phase_3": {
        "percentage": 50,
        "rules": [
          {
            "name": "random_50_percent",
            "percentage": 50,
            "variation": true
          }
        ]
      },
      "phase_4": {
        "percentage": 100,
        "rules": [
          {
            "name": "all_users",
            "percentage": 100,
            "variation": true
          }
        ]
      }
    }
  }
}

#### 2.2 Flag状态追踪

状态记录：

{
  "flag_status_tracking": {
    "flag_key": "feature_wechat_login",
    "current_phase": "phase_2",
    "current_traffic_percentage": 10,
    "last_updated": "ISO8601",
    "updated_by": "release-gradual-pipeline",
    "history": [
      {"timestamp": "ISO8601", "phase": "phase_1", "percentage": 1, "action": "enabled"},
      {"timestamp": "ISO8601", "phase": "phase_2", "percentage": 10, "action": "increased"},
      {"timestamp": "ISO8601", "phase": "phase_2", "percentage": 0, "action": "rollback"}
    ]
  }
}

Step 3: 各阶段自动监控

#### 3.1 指标采集

采集配置：

{
  "metrics_collection": {
    "interval_seconds": 30,
    "metrics": [
      {"name": "error_rate", "source": "prometheus", "query": "rate(http_requests_total{status=~'5..'}[5m])"},
      {"name": "latency_p99", "source": "prometheus", "query": "histogram_quantile(0.99, rate(http_request_duration_seconds_bucket[5m]))"},
      {"name": "availability", "source": "synthetic_monitoring", "frequency": "1m"}
    ]
  }
}

#### 3.2 阶段状态评估

评估规则：

指标类型	评估条件	评估结果
P0全部正常	无指标超warning阈值	可进入下一阶段
P0任一告警	任一指标超warning但未超rollback	继续观察
P0恶化	任一指标超rollback阈值	立即回滚
P1告警	P1指标超阈值	延迟+告警

阶段评估输出：

{
  "phase_assessment": {
    "phase_id": "phase_2",
    "assessed_at": "ISO8601",
    "duration_minutes": 90,
    "metrics_summary": {
      "p0_metrics": {
        "error_rate": {"current": 0.002, "baseline": 0.001, "status": "normal"},
        "latency_p99": {"current": 150, "baseline": 140, "status": "normal"},
        "availability": {"current": 99.99, "baseline": 99.99, "status": "normal"}
      },
      "p1_metrics": {
        "api_success_rate": {"current": 99.5, "baseline": 99.8, "status": "normal"}
      },
      "p2_metrics": {
        "conversion_rate": {"current": -2, "baseline": 0, "status": "normal"}
      }
    },
    "overall_status": "healthy",
    "recommendation": "can_proceed",
    "confidence": 0.95
  }
}

#### 3.3 阶段转换决策

决策流程：

┌─────────────────┐
│ 阶段时间已满？  │
└────────┬────────┘
         │
    ┌────┴────┐
    │         │
   是        否
    │         │
    ▼         ▼
┌────────┐ ┌─────────────────┐
│指标评估│ │继续监控         │
└───┬────┘ └─────────────────┘
    │
    ▼
┌─────────────────┐
│ 所有P0指标正常？│
└────────┬────────┘
         │
    ┌────┴────┐
    │         │
   是        否
    │         │
    ▼         ▼
┌────────┐ ┌─────────────────┐
│进入下一 │ │ 告警/观察/回滚  │
│阶段     │ └─────────────────┘
└────────┘

转换输出：

{
  "phase_transition_decision": {
    "current_phase": "phase_2",
    "decision": "proceed_to_next_phase",
    "next_phase": "phase_3",
    "reason": "阶段持续时间已满，所有P0指标正常",
    "metrics_verified": true,
    "approved_by": "release-gradual-pipeline",
    "scheduled_at": "ISO8601"
  }
}

Step 4: 自动回滚触发

#### 4.1 回滚触发条件

触发规则：

条件	触发动作	延迟
P0指标超rollback阈值	立即自动回滚	0秒
服务不可用	立即自动回滚	0秒
P1指标超rollback阈值	延迟观察	5分钟
人工确认异常	手动触发	0秒

#### 4.2 回滚执行

回滚流程：

触发回滚
    │
    ▼
┌─────────────────┐
│ 通知相关人员    │
└────────┬────────┘
         │
    ┌────┴────┐
    │         │
   自动化    人工
    │         │
    ▼         ▼
┌─────────┐ ┌─────────────────┐
│关闭Flag │ │确认并执行      │
│切换流量 │ └─────────────────┘
└────┬────┘
     │
     ▼
┌─────────────────┐
│ 验证回滚结果    │
└────────┬────────┘
         │
    ┌────┴────┐
    │         │
   成功      失败
    │         │
    ▼         ▼
┌─────────┐ ┌─────────────────┐
│记录完成 │ │告警升级人工介入 │
└─────────┘ └─────────────────┘

回滚执行输出：

{
  "rollback_execution": {
    "rollback_id": "rollback_2024_0125_002",
    "trigger": {
      "reason": "P0指标恶化",
      "metric": "error_rate",
      "current_value": 0.035,
      "threshold": 0.03
    },
    "action": {
      "step_1": "关闭Feature Flag",
      "step_2": "等待流量完全切换",
      "step_3": "验证旧版本健康"
    },
    "status": "completed",
    "completed_at": "ISO8601",
    "duration_seconds": 45,
    "verification": {
      "traffic_restored": true,
      "old_version_healthy": true,
      "error_rate_recovered": true
    }
  }
}

#### 4.3 回滚后处理

处理清单：

{
  "post_rollback_actions": [
    {
      "action": "通知发布团队",
      "type": "notification",
      "recipients": ["release_lead", "dev_lead", "product_manager"]
    },
    {
      "action": "记录回滚事件",
      "type": "documentation",
      "content": "回滚原因、持续时间、影响范围"
    },
    {
      "action": "创建事件跟进Ticket",
      "type": "ticket",
      "template": "post_incident_review"
    },
    {
      "action": "分析根因",
      "type": "analysis",
      "deadline": "24小时内"
    }
  ]
}

输出

存储路径：output/pm-monitoring/release-gradual/

输出文件：release_status.json

输出Schema：

{
  "type": "object",
  "required": ["output_id", "release_id", "release_status"],
  "properties": {
    "output_id": {"type": "string", "description": "输出唯一标识"},
    "release_id": {"type": "string", "description": "发布ID"},
    "generated_at": {"type": "string", "description": "生成时间"},
    "release_status": {"type": "object", "description": "当前发布状态，包含阶段、流量比例和进度"},
    "phase_transitions": {"type": "array", "description": "阶段转换历史记录"},
    "rollback_history": {"type": "array", "description": "回滚历史记录"},
    "monitoring_metrics": {"type": "object", "description": "监控数据汇总，包含当前和历史指标"},
    "canary_plan": {"type": "object", "description": "灰度发布计划"}
  }
}

最终输出结构

{
  "output_id": "release_status_xxx",
  "release_id": "release_2024_0125_001",
  "generated_at": "ISO8601",
  "release_status": {
    "current_phase": "phase_3",
    "traffic_percentage": 50,
    "start_time": "ISO8601",
    "elapsed_minutes": 180,
    "status": "in_progress"
  },
  "phase_transitions": [
    {
      "phase_id": "phase_1",
      "traffic_percentage": 1,
      "started_at": "ISO8601",
      "completed_at": "ISO8601",
      "duration_minutes": 32,
      "outcome": "promoted",
      "metrics_at_completion": {...}
    },
    {
      "phase_id": "phase_2",
      "traffic_percentage": 10,
      "started_at": "ISO8601",
      "completed_at": "ISO8601",
      "duration_minutes": 125,
      "outcome": "promoted",
      "metrics_at_completion": {...}
    }
  ],
  "rollback_history": [],
  "monitoring_metrics": {
    "current": {...},
    "historical": [...]
  },
  "canary_plan": {...}
}

输出字段说明

字段	类型	说明
release_status	JSON	当前发布状态
phase_transitions	JSON	阶段转换历史
rollback_history	JSON	回滚历史
monitoring_metrics	JSON	监控数据汇总

输出校验规则

字段路径	类型	必填	说明
gradual_release	object	是	灰度发布根对象
gradual_release.strategy	object	是	灰度策略
gradual_release.strategy.type	string	是	策略类型，枚举值：canary/blue_green/rolling/feature_flag
gradual_release.strategy.stages	array	是	灰度阶段列表，至少2个
gradual_release.strategy.stages[].name	string	是	阶段名称
gradual_release.strategy.stages[].traffic_percentage	number	是	流量百分比，0-100
gradual_release.strategy.stages[].duration	string	是	持续时间
gradual_release.strategy.stages[].success_criteria	object	是	成功标准
gradual_release.strategy.stages[].rollback_criteria	object	是	回滚标准
gradual_release.monitoring	object	是	监控配置
gradual_release.monitoring.metrics	array	是	监控指标列表
gradual_release.monitoring.alert_rules	array	是	告警规则列表
gradual_release.rollback_plan	object	是	回滚计划
gradual_release.rollback_plan.trigger_conditions	array	是	触发条件
gradual_release.rollback_plan.steps	array	是	回滚步骤

上游变更响应

当上游输入发生变更时，本Skill的响应策略：

上游变更	影响范围	响应策略
发布检查清单变更	灰度策略和门禁	重新评估灰度门禁条件，标记需人类确认
验收报告变更	灰度启动条件	更新灰度启动条件，重新评估是否可以开始灰度
监控指标变更	监控配置和告警规则	更新监控指标和告警阈值，标记需人类确认
PRD需求变更	灰度范围和功能验证	重新评估灰度范围，标记需人类确认

当灰度发布自身变更时，对下游的通知机制：

灰度变更类型	通知范围	通知方式
灰度阶段推进	release-notes	标记阶段推进，触发发布说明更新
灰度回滚	全部下游	标记回滚事件，触发问题排查和修复
灰度完成	release-notes	标记灰度完成，触发全量发布决策

决策规则

自动回滚规则

触发条件	执行动作	优先级
P0指标超rollback阈值	立即自动回滚	最高
服务完全不可用	立即自动回滚	最高
P1指标超rollback阈值	5分钟延迟后回滚	高
P2指标超rollback阈值	告警，等待人工确认	中

阶段通过规则

条件	决策
P0指标全部正常	可进入下一阶段
阶段持续时间已满	评估是否进入下一阶段
存在P0告警（未超阈值）	延长观察时间
存在P1告警	警告，进入下一阶段

质量检查

Quality Gates

检查项	标准	未达标处理
发布前验证	所有验证项通过	阻止发布
P0指标稳定	指标无恶化趋势	停止灰度
回滚机制就绪	Feature Flag可用	阻止发布
告警配置正确	所有指标已配置告警	阻止发布

质量检查清单

[ ] 发布前验证全部通过
[ ] Feature Flag配置正确
[ ] 监控告警已配置
[ ] 回滚机制就绪
[ ] 值班人员已确认
[ ] 阶段转换记录完整

降级策略

上游文件缺失降级方案

缺失范围	降级方案	输出影响
发布内容缺失	用户提供发布功能列表 → 生成灰度计划模板	发布内容细节需人工补充
Feature Flag配置缺失	生成灰度计划但不包含Flag自动配置步骤	需人工配置Feature Flag
监控指标缺失	使用默认P0/P1/P2指标模板，标注"待确认"	告警阈值基于通用模板
发布内容 + Feature Flag + 监控指标均缺失	用户提供发布功能列表 → 生成灰度计划模板	输出灰度计划模板，各配置项标注"待确认"

数据获取说明

当上游文件缺失时，需用户提供以下信息以支撑降级生成：

发布功能列表：本次发布包含的功能和变更
回滚版本（可选）：可回滚的上一版本号
关键监控指标（可选）：需要重点关注的业务和技术指标

执行日志

{
  "execution_id": "exec_p6_xxx",
  "pipeline": "release-gradual",
  "release_id": "release_2024_0125_001",
  "trigger": "quality_gate_passed",
  "started_at": "ISO8601",
  "completed_at": "ISO8601",
  "phases": [
    {"phase_id": "phase_1", "status": "completed", "duration_minutes": 32, "outcome": "promoted"},
    {"phase_id": "phase_2", "status": "completed", "duration_minutes": 125, "outcome": "promoted"},
    {"phase_id": "phase_3", "status": "in_progress", "duration_minutes": 45}
  ],
  "rollbacks": [],
  "final_status": {
    "outcome": "in_progress",
    "current_traffic": 50,
    "metrics_healthy": true
  }
}

release-gradual— agent skill

Is release-gradual safe to install?

灰度发布自动执行

核心原则

交互模式

输入

发布内容结构示例

灰度阶段配置

标准灰度阶段

护栏指标配置

执行步骤

Step 1: 灰度计划生成

Step 2: Feature Flag自动配置

Step 3: 各阶段自动监控

Step 4: 自动回滚触发

输出

最终输出结构

输出字段说明

输出校验规则

上游变更响应

决策规则

自动回滚规则

阶段通过规则

质量检查

Quality Gates

质量检查清单

降级策略

上游文件缺失降级方案

数据获取说明

执行日志

Audit the pieces. Scan the whole. Decide.