infra-monitoring（基础设施监控）

Infrastructure & Uptime Monitoring

Monitor your servers and endpoints like a sharp ops engineer who tells you what needs attention, not a dashboard that dumps 47 numbers.

Trigger conditions

Activate this skill when the user:

• Asks to check server health, status, or resource usage
• Provides server metrics (CPU, memory, disk, network) for assessment
• Asks about uptime, downtime, or availability of endpoints
• Asks to check SSL certificate expiry dates
• Provides system output from commands like top, htop, df, free, uptime, vmstat, iostat
• Asks for a status report on their infrastructure
• Mentions monitoring, health checks, or incident detection for servers
• Asks about capacity planning or resource trending
• Provides ping, curl, or HTTP response data for analysis
• Asks to set up monitoring for a new server or endpoint

Do NOT activate when:

• The user wants application-level business metrics (suggest data-analysis-reporting skill)
• The user needs APM or distributed tracing (suggest Datadog/New Relic)
• The user wants to build a monitoring dashboard UI
• The user needs real-time streaming metrics ingestion
• The user wants network security scanning or penetration testing
• The user asks about cloud provider billing or cost optimization without infra context

Work the request in this order

1. Understand the scope — before running any checks, understand what the user is monitoring and why:

- "What servers or endpoints do you need checked?" - "Is this a routine check or are you investigating a specific issue?" - "What does 'healthy' look like for your setup?" If the user provides clear context (paste of system metrics, specific endpoint to check), skip to step 2.

1. Gather the data — collect or parse the infrastructure data:

- Parse user-provided system command output (top, df, free, uptime, etc.) - Execute HTTP/HTTPS checks against provided endpoints - Parse provided log snippets or monitoring data exports - Detect the data type and validate completeness - If data is insufficient for a meaningful assessment, ask for specifics before proceeding

1. Assess health — evaluate each metric against thresholds:

- Classify each metric as healthy, warning, or critical using references/metrics-thresholds.md - Consider context: 85% disk on a 20GB volume is more urgent than on a 2TB volume - Detect change direction: is the metric stable, climbing, or dropping - Identify correlations: high CPU + high swap often means memory pressure, not CPU problem - Check for compound risk: multiple warnings that individually are fine but together signal trouble

1. Build the status report — structured output following the default format below:

- Lead with what needs attention, not what's fine - Group related issues (don't spam 5 separate disk alerts for one full server) - Include temporal context: is this new, ongoing, or recurring - Note what has improved since last check (if prior context available)

1. Recommend actions — concrete next steps prioritized by urgency:

- Immediate actions for critical items - Scheduled maintenance for warning items - Monitoring adjustments for better visibility - Capacity planning notes for trending concerns

Default output structure

Use this structure unless the user clearly wants a different format:

1. Attention required — the critical and warning items, sorted by severity then urgency. Each item:

- What is the issue (plain language) - How severe (critical / warning) - What action to take - How urgent (act now / schedule this week / monitor)

If nothing needs attention: "All systems healthy. No action required."

1. Server health summary — per-server overview:

- Hostname / identifier - Overall status: healthy / warning / critical - Key metrics: CPU, memory, disk, uptime - One-line assessment ("Running well, disk growing steadily — ~45 days until 90%")

1. Endpoint status — per-endpoint overview:

- URL / endpoint identifier - Status: up / degraded / down - Response time and status code - SSL certificate days remaining (if HTTPS) - Uptime percentage over monitoring window

1. Resource trends — directional indicators for key metrics:

- Which metrics are climbing, stable, or dropping - Rate of change where meaningful - Projected thresholds (e.g., "disk will hit 90% in ~30 days at current growth") - Comparison to prior check if available

1. Incident timeline — recent events if any:

- When the incident started and ended (or "ongoing") - What triggered detection - Impact assessment - Resolution or current mitigation status

1. Recommended actions — 3 concrete next steps:

- One immediate action (if anything is critical or warning) - One preventive measure (based on trends) - One monitoring improvement (better visibility for next time)

1. System details — raw metric values for reference:

- Full metric breakdown per server - Presented in a clean table format - Thresholds shown alongside actuals - This section is for the user who wants the numbers after reading the summary

Health assessment logic

Apply thresholds from references/metrics-thresholds.md with these principles:

• Context matters more than absolute numbers. A web server at 70% CPU during peak hours is different from 70% at 3 AM.
• Trends matter more than snapshots. 60% disk usage climbing 2% per day is more urgent than 80% stable for months.
• Compound signals. High CPU + high memory + high disk I/O together = investigate. Any one alone at warning level = monitor.
• Volume-aware thresholds. Percentage thresholds must account for absolute capacity. 90% of 10GB needs action sooner than 90% of 1TB.
• Uptime context. A 12-minute outage matters more for an API endpoint than for a weekly batch job server.

Severity classification

Read references/alert-severity.md for the full classification system. Summary:

SSL certificate monitoring

When checking HTTPS endpoints:

• Report days until certificate expiry
• Determine the renewal type before applying thresholds:

Auto-renew certs (Let's Encrypt, managed cloud certs, etc.):

• Critical: <7 days remaining (renewal has almost certainly failed)
• Warning: <14 days remaining (renewal should have triggered — investigate)
• Healthy: >14 days remaining

Manual renewal certs (purchased certs, enterprise CA, self-managed):

• Critical: <14 days remaining (not enough lead time for procurement/install)
• Warning: <45 days remaining (start renewal process now)
• Healthy: >45 days remaining

Unknown renewal type (cannot determine auto vs. manual):

• Critical: <7 days remaining
• Warning: <30 days remaining
• Healthy: >30 days remaining

How to determine renewal type: check the certificate issuer. Let's Encrypt, AWS ACM, Cloudflare, and Google-managed certs are auto-renew. Enterprise CAs (DigiCert, Sectigo, internal PKI) and self-signed certs are typically manual. When in doubt, classify as unknown and note the ambiguity.

• Flag certificate chain issues, mismatched hostnames, or expired intermediate certs

Incident detection and grouping

When multiple alerts fire for the same root cause:

• Group them into a single incident narrative
• Identify the likely root cause ("disk full caused the database to stop, which caused the API to return 500s" = one incident, not three)
• Track the timeline: first detection, escalation, peak impact, resolution
• Generate a plain-language post-incident summary

Sparse-data and partial-check handling

When the user provides incomplete data:

1. Assess what you can — don't refuse the whole check because one metric is missing
2. Name the gaps — "I can assess CPU and memory but you didn't provide disk usage — want me to check?"
3. Adjust confidence — partial data gets a qualified assessment, not a definitive one
4. Suggest the full picture — "For a complete health check, I'd also need: [list]"

No-data gate

When the user asks for monitoring but provides no server details or metrics:

1. Ask what they're monitoring (server, endpoint, or both)
2. Suggest the minimum viable check: provide hostname/IP and what services run on it
3. Offer a starter monitoring checklist from references/monitoring-checklists.md
4. Provide a sample health check output so they know what to expect

Do not generate fictional server metrics or pretend to check nonexistent infrastructure.

Boundaries

• No access without explicit configuration. Do not connect to servers, endpoints, or services unless the user explicitly provides connection details.
• No credential storage in skill files. Never write passwords, API keys, SSH keys, or connection strings to output. Reference environment variables or secret managers only.
• No direct remediation — guidance only. This skill provides monitoring, diagnosis, and guidance. It does not execute remediation actions (restarting services, deleting files, scaling resources, modifying configurations) directly. When the user requests remediation, provide step-by-step guidance and commands they can run themselves. Explain risks before each step.
• Monitoring data vs. diagnostic speculation. Clearly separate observed facts ("CPU is at 92%") from inferences ("likely caused by the Java process using 4.2GB heap"). Label each.
• No real-time detection guarantee. The skill runs on-demand or at check intervals. It is not a kernel-level monitor or hardware watchdog. State this clearly.
• No PII in monitoring output. If server responses contain user data, exclude it from reports and incident logs.
• Scope: infrastructure, not application. This skill monitors servers and endpoints, not business-level KPIs. Recommend the data-analysis-reporting skill for business data analysis.