菜单
Instrument services and triage errors. Trigger with "why is this slow", "what's erroring", "set up alerting", "add tracing", "define an SLO", "triage this Sentry issue", "we're getting paged for nothing", or when the user needs logs, metrics, traces, or error monitoring.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Atlas architect: boot the workspace. Verify claude-mem and context-mode, scan the project, recommend tooling (confirm first), wire hooks, seed the docs/ SSOT.
Measure atlas's own run health, audit context/asset waste, and mine session transcripts. Three lenses: RUN HEALTH reads the observability DB for metrics (wall-clock, dispatches, verifier coverage) and baseline->target goals; ASSET AUDIT scores each context-loaded skill/agent/plugin by token cost and relevance to flag disable/relocate; SESSION FORENSICS mines transcripts for unused tools and unverified claims into rules. No args reports trends.
Use to map a codebase into feature-grouped flowcharts, find architectural duplication across features, and propose the simplest unified architecture - discovery-first and zero-arg. Runs as a Workflow that fans out one explorer per feature, hunts duplication, and synthesizes a unification proposal with file:line evidence. Use before a refactor, to "find the ideal path," or to unify duplicated systems.
Generate or improve a company-specific data analysis skill by extracting tribal knowledge from analysts. BOOTSTRAP MODE - Triggers: "Create a data context skill", "Set up data analysis for our warehouse", "Help me create a skill for our database", "Generate a data skill for [company]" -> Discovers schemas, asks key questions, generates initial skill with reference files ITERATION MODE - Triggers: "Add context about [domain]", "The skill needs more info about [topic]", "Update the data skill with [metrics/tables/terminology]", "Improve the [domain] reference" -> Loads existing skill, asks targeted questions, appends/updates reference files Use when data analysts want Claude to understand their company's specific data warehouse, terminology, metrics definitions, and common query patterns.
QA an analysis before sharing with stakeholders - methodology checks, accuracy verification, and bias detection. Use when reviewing an analysis for errors, checking for survivorship bias, validating aggregation logic, or preparing documentation for reproducibility.
Build self-contained interactive HTML dashboards with Chart.js, dropdown filters, and professional styling. Use when creating dashboards, building interactive reports, or generating shareable HTML files with charts and filters that work without a server.
| name | observability |
| description | Instrument services and triage errors. Trigger with "why is this slow", "what's erroring", "set up alerting", "add tracing", "define an SLO", "triage this Sentry issue", "we're getting paged for nothing", or when the user needs logs, metrics, traces, or error monitoring. |
Make systems explainable: instrument them, then triage what they report.
| Pillar | Answers | Use For |
|---|---|---|
| Logs | What happened, in detail | Forensics, audit trail, debugging one request |
| Metrics | How much, how often, how fast | Trends, dashboards, alert thresholds |
| Traces | Where the time went across services | Latency breakdown, "why is this slow" |
| Errors | What broke, how often, for whom | Triage, regression detection, release health |
Logs and traces explain a single request. Metrics and errors explain the aggregate. Reach for the matching pillar before guessing.
Emit JSON, not free text. Every log line carries a correlation ID (request or trace ID) so one request can be reconstructed across services. Propagate that ID through every hop and attach it to errors and spans. Redact secrets and PII at the logger, not at the call site. Log the why and the context, not a restatement of the line of code.
Work the funnel in order. Do not jump to root cause before scoping impact.
Alert on symptoms users feel (error rate, latency, availability), not on causes (CPU, memory, queue depth). Cause metrics belong on dashboards for diagnosis, not in the pager.
