AgentSkillsCN

Capacity Planning

基于增长预测，合理预估资源需求并设计扩容策略。

SKILL.md
--- frontmatter
name: Capacity Planning
description: Forecast resource needs and design scaling strategies based on growth projections
category: operations
version: 1.0.0
triggers:
  - utilization-threshold
  - quarterly-review
  - scaling-strategy
globs: "**/infrastructure/**,**/scaling/**"

Capacity Planning Skill

Forecast resource needs and design scaling strategies based on growth projections.

Trigger Conditions

  • •Traffic spikes or utilization exceeds 75% threshold
  • •Quarterly capacity review cycle
  • •User invokes with "capacity plan" or "scaling strategy"

Input Contract

  • •Required: Current resource utilization metrics
  • •Required: Growth forecast (user acquisition, traffic patterns)
  • •Optional: Budget constraints, SLA requirements

Output Contract

  • •Capacity model with projections
  • •Scaling recommendations (horizontal/vertical, auto/manual)
  • •Cost forecast per growth scenario
  • •Headroom analysis for failover scenarios

Tool Permissions

  • •Read: Metrics, infrastructure configs, billing data
  • •Write: Capacity planning documents
  • •Search: Historical usage patterns

Execution Steps

  1. •Collect current utilization across compute, storage, network
  2. •Model demand from business metrics, not just historical extrapolation
  3. •Define target utilization bands (60-75%)
  4. •Calculate headroom for failover (lose 1 of N replicas)
  5. •Project costs for 3, 6, 12 month horizons
  6. •Recommend scaling triggers and strategies
  7. •Document the capacity model

Success Criteria

  • •Utilization targets defined per resource type
  • •Cost projections for multiple growth scenarios
  • •Scaling triggers calibrated for scale-up latency
  • •Failover headroom verified

Escalation Rules

  • •Escalate if projected costs exceed budget by >20%
  • •Escalate if current utilization exceeds 85%
  • •Escalate if no scaling strategy can meet SLA targets

Example Invocations

Input: "Plan capacity for Black Friday (expected 5x normal traffic)"

Output: Current: 3 pods, 45% CPU avg. For 5x: need 12 pods minimum (15 recommended for headroom). Auto-scaler trigger at 60% CPU. Pre-scale 2 hours before peak. Estimated additional cost: $2,400 for 48h burst. DB read replicas: add 2 in us-east, 1 in eu-west.