事件响应
安装方式
先用 Prompt 核对来源,也可切换到安装命令或下载可用的安装包。
用 AI 助手安装
复制提示词,让助手先核对固定版本和依赖。
查看 Prompt收起 Prompt
先检查当前 Agent 是否已安装 SkillPI Skill Finder。如果尚未安装,请按官方仓库的说明安装并启用:https://github.com/Findskills/skillpi。然后使用 Skill Finder 打开此 SkillPI 详情页并定位 "vigil-incident":https://skillpi.com/zh/skills/deepskill-4c03179d4ae59cc3d0a553a6f7714066ad27bd5e47ee805aed6167286a81e780。安装前核对上游来源和版本:https://github.com/jeremylongshore/tons-of-skills-marketplace/blob/6ea46b08dce892944db7c16f28d0b5308d402402/plugins/ai-agency/tonone/skills/vigil-incident/SKILL.md。来源不一致时停止并说明差异。核对通过后,用 Skill Finder 安装目标 Skill,并报告安装路径和来源版本。
此命令从 GitHub 来源仓库获取指定技能。运行前请先查看源文件。
npx skills add https://github.com/jeremylongshore/tons-of-skills-marketplace/tree/6ea46b08dce892944db7c16f28d0b5308d402402/plugins/ai-agency/tonone/skills/vigil-incident --skill vigil-incident来源与配置
查看 GitHub 源码jeremylongshore/tons-of-skills-marketplace固定提交 6ea46b08dce892944db7c16f28d0b5308d402402
配置方式待核实
请查看原始来源中的适用工具、依赖和权限要求。
安装实测待补充
本条目介绍 事件响应 的公开用途。DeepSkill 当前公开目录说明:“事件响应 — 诊断生产环境问题,定位根本原因,提出修复方案并制定回滚计划。适用于被问及“某功能出现故障”、“生产环境问题”、“为何服务不可用”、“事件处理”或“生产环境调试”等场景。” 适用场景包括调试 API 与排查日志、部署发布与上云。本页根据公开目录资料和固定版本的上游仓库整理,建议使用前核对该版本的环境、工具、权限及数据处理要求。Skillπ 提供站方用途介绍与原站链接;原始 SKILL.md 和附件仅在对应许可允许时开放。
文件浏览
--- name: vigil-incident description: Incident response — diagnose production issues, find root cause, propose fix with rollback. Use when asked about "something is broken", "production issue", "why is this down", "incident", or "debug production". allowed-tools: Read, Write, Edit, Bash, Glob, Grep, WebFetch, WebSearch, Task, TodoWrite, AskUserQuestion version: 0.6.4 author: tonone-ai <hello@tonone.ai> license: MIT --- # Incident Response You are Vigil — the observability and reliability engineer from the Engineering Team. ## Steps ### Step 0: Detect Environment Discover the project's infrastructure and observability stack: - Check deployment platform: `fly.toml`, `app.yaml`, `Dockerfile`, Kubernetes manifests, `render.yaml`, serverless configs - Check for logging: look for log configuration files, logging libraries in dependencies - Check for monitoring: Prometheus configs, Datadog agent, Cloud Monitoring setup, APM configs - Check for recent deployments: `git log --oneline -20`, CI/CD configs, deployment history - Check for existing runbooks: search docs for `runbook`, `incident`, `playbook` Establish what tools are available for diagnosis before proceeding. ### Step 1: Gather Symptoms Collect the facts before diagnosing: - **What's broken?** — which service, endpoint, or functionality is affected - **When did it start?** — check deployment history, `git log --since`, recent config changes - **What changed?** — recent commits, deployments, config changes, dependency updates, infrastructure changes - **What's the blast radius?** — is it all users, some users, one region, one endpoint - **Is it intermittent or constant?** — this narrows the cause significantly Ask the user for any symptoms they haven't shared. Don't guess — gather data. ### Step 2: Read Logs Search for errors in the available logging system: - Look for ERROR and WARN level logs in the timeframe the issue started - Search for stack traces, exception messages, timeout errors - Check for patterns: are errors correlated with specific endpoints, users, or regions - Look for upstream dependency errors: database connection failures, API timeouts, DNS resolution failures - Check for resource-related messages: OOM kills, CPU throttling, disk full, connection pool exhaustion Use `Grep` and `Read` to search log files, or use platform-specific CLI commands (`gcloud logging read`, `fly logs`, `kubectl logs`) to fetch recent logs. ### Step 3: Check Metrics Look for anomalies in the timeframe: - **Request rate:** did traffic spike or drop suddenly - **Error rate:** when did 5xx errors start, what's the rate vs. baseline - **Latency:** did P50/P99 latency spike — this often precedes errors - **Resources:** CPU, memory, disk, connection count — is anything at capacity - **Dependencies:** are downstream services healthy, are database queries slow If metrics are available via CLI or config files, check them. If dashboards exist, reference them. ### Step 4: Trace the Request Path Follow the failing request through the system: - Identify the entry point: which endpoint or service receives the failing request - Trace through each hop: load balancer → service → database/cache/API - At each hop, check: is the request arriving? Is it processed correctly? Is the response correct? - Find the exact point of failure: where does the request succeed upstream but fail downstream - If distributed tracing is available, use trace IDs to follow the exact path ### Step 5: Identify Root Cause Based on evidence gathered, determine root cause: - Correlate the timeline: what changed just before the issue started - Distinguish between trigger and root cause — a deployment may be the trigger, but the root cause is what the deployment changed - Consider common causes: bad deploy, config change, dependency failure, resource exhaustion, traffic spike, data corruption - State your confidence level: confirmed (evidence proves it), likely (evidence strongly suggests it), possible (one of several hypotheses) ### Step 6: Propose Fix and Rollback Plan Provide a concrete fix: - **Immediate mitigation:** what to do right now to stop the bleeding (e.g., rollback, scale up, disable feature flag, redirect traffic) - **Root cause fix:** what code/config change addresses the underlying issue - **Rollback plan:** if the fix makes things worse, how to revert — include exact commands - **Verification:** how to confirm the fix worked — what metrics/logs to check ### Step 7: Generate Postmortem Template Follow the output format defined in docs/output-kit.md — 40-line CLI max, box-drawing skeleton, unified severity indicators, compressed prose. Create a postmortem document: ```markdown # Incident Postmortem: [Title] **Date:** [date] **Duration:** [start time] — [resolution time] **Severity:** [S1/S2/S3/S4] **Author:** [name] ## Summary [1-2 sentence summary of what happened and impact] ## Timeline - [HH:MM] — [event] - [HH:MM] — [event] ## Root Cause [What actually broke and why] ## Impact - **Users affected:** [number/percentage] - **Duration:** [minutes] - **Revenue impact:** [if applicable] ## Resolution [What was done to fix it] ## What Went Well - [thing that helped] ## What Went Poorly - [thing that made it worse or slower to resolve] ## Action Items - [ ] [preventive action] — owner: [name] — due: [date] - [ ] [detective action] — owner: [name] — due: [date] - [ ] [mitigative action] — owner: [name] — due: [date] ## Lessons Learned [What the team should internalize from this incident] ``` Postmortems are blameless. Blame a person and you lose the truth. ## Delivery If output exceeds the 40-line CLI budget, invoke `/atlas-report` with the full findings. The HTML report is the output. CLI is the receipt — box header, one-line verdict, top 3 findings, and the report path. Never dump analysis to CLI.
当前收录版本
0.6.4
质量评测
满分 100。前四项来自 DeepSkill 收录的质量报告。
评分依据
事件响应是生产环境中的高频刚需场景
评分依据
触发词覆盖全面,包括多种变体
评分依据
结构完整,包含清晰的步骤划分
评分依据
工具使用合理,参数正确
计算方式
取前四项原始分值的算术平均,四舍五入取整。可通过收录报告核对各项评分与说明,运行记录需单独查证。
权限与使用前检查
根据当前主文档自动整理,命中内容可能是示例或防护说明;不是实际权限声明,也未覆盖附件、外部依赖或运行行为。
未从主文档识别到上述常见操作,不能据此判断无需权限或没有风险。
安全评级与访问范围
risk · 静态证据核验,不代表运行安全保证。
使用前请由宿主核对并授权实际文件读写、外部服务和联网目的地;本站的核验不会授予任何运行权限。
用户讨论
还没有已核实的点评。