Cost optimization

cost-optimization

Use this skill to work through cost optimization recommendations from discovery to remediation tracking.

Workflow

Choose the entry path that matches the user's object:

1. Analyze an existing recommendation:
   • List recommendations with list_recommendations.py
   • Optional: --with-related-policies to show related policy counts
   • Analyze a recommendation with analyze_recommendation.py
   • Silently resolve the related resourceId through datasource ../datasource/scripts/list_resource.py
   • Merge normalized resource type + properties into the analysis facts
   • Returns multi-dimensional recommendations (P0/P1/P2 priority)
   • Includes risk assessment and best practice guidance
   • Shows saving contribution, policy history, and resource operational context
2. Analyze a resource directly:
   • Call analyze_resource_cost.py with an exact visible name or recent list # selection
   • Read resource facts, enabled applicable policy configurations, latest resource executions, and active violations without triggering policy execution
   • Use the returned analysisContract to keep platform facts separate from llm_potential
   • Read references/RESOURCE_ANALYSIS.md for VM, AWS RDS, and generic resource reasoning rules
3. Remediate an existing finding through the native day2 repair in execute_optimization.py only after the user explicitly requests it
4. Track remediation state with track_execution.py

Analysis Output Enhancement

The analyze_recommendation.py now provides:

• P0 Primary Action: Core recommendation (remediate / configure_platform_policy / manual_review)
• P1 Risk Assessment: Risk level (high/medium/low) with specific warnings
• P1 Configuration Guide: When fixType is missing, explains how to configure day2 repair
• P1 Saving Priority: Contribution percentage to global optimizable amount
• P2 Policy History: Compliance rate trend and violation recurrence count
• Resource Context: Resource type, component type, status, OS, and normalized datasource facts

Safety Boundary

The skill only performs platform-native remediation through:

• POST /compliance-policies/violations/day2/fix/{id}

It does not call AWS or Azure APIs directly.

Resource-first analysis is read-only. It must not call:

• POST /compliance-policies/execute
• POST /compliance-policies/violations/day2/fix/{id}

Only an existing platform violation may enter the separate remediation flow. An llm_potential result is never executable.

Resource Enrichment

This skill should internally reuse the datasource skill's shared ../datasource/scripts/list_resource.py helper whenever a recommendation includes resourceId.

• Pull resource details before rendering the final analysis output.
• Merge resource status/type/OS and normalized facts into facts and downstream recommendations.
• If resource lookup is unavailable, continue with policy/violation analysis as a best-effort degradation path.