terraform-plan-review（Terraform plan 审查）

Terraform Plan Review

Overview

Analyze terraform plan output using parallel agents for comprehensive risk assessment. Never auto-apply - always present findings and require explicit approval.

Announce at start: "I'm using the terraform-plan-review skill to analyze these changes safely."

The Process

Step 1: Verify Environment

Before running any plan:

1. Check AWS Profile aws sts get-caller-identity

- Verify the account ID matches expected environment - Verify the role/user is appropriate for this operation - If mismatch: STOP and alert user

1. Identify Environment

- Check current directory structure (which environment?) - Verify backend configuration matches environment

Step 2: Generate Plan

# Initialize if needed
terraform init

# Generate plan file (required for JSON parsing)
terraform plan -out=plan.out

# Convert to JSON for analysis
terraform show -json plan.out > plan.json

Step 3: Dispatch Parallel Analysis Agents

Launch these agents in a single message with multiple Task calls:

Task 1:
  description: "Analyze plan risks"
  prompt: |
    Analyze this Terraform plan for risks and impact.
    Environment: [env name]
    Account: [account id]

    Plan JSON:
    [plan.json content]

    Focus on destruction, modification risks, and cascade effects.
  subagent_type: "terraform-plan-analyzer"

Task 2:
  description: "Security review plan"
  prompt: |
    Review this Terraform plan for security implications.
    Environment: [env name]

    Plan JSON:
    [plan.json content]

    Focus on IAM, network, encryption, and compliance.
  subagent_type: "security-reviewer"

Task 3:
  description: "Check historical patterns"
  prompt: |
    Analyze git history for patterns related to these resources.
    Resources being changed: [list from plan]

    Look for similar past changes, incidents, and outcomes.
  subagent_type: "historical-pattern-analyzer"

CRITICAL: All three Task calls in ONE message for parallel execution.

Agent prompts should include:

• The plan.json content (or path)
• The environment name
• Any relevant context from memory

Step 4: Aggregate Findings

Collect results from all agents and create a unified report:

## Plan Analysis Summary

### Risk Level: [CRITICAL/HIGH/MEDIUM/LOW]

### Changes Overview
- Resources to create: X
- Resources to update: Y
- Resources to destroy: Z

### Risk Analysis (terraform-plan-analyzer)
[Summary of risks identified]

### Security Analysis (security-reviewer)
[Summary of security implications]

### Pattern Analysis (historical-pattern-analyzer)
[Any similar past changes and their outcomes]

### Required Approvals
- [ ] User acknowledges destruction of X resources
- [ ] User confirms this is the correct environment
- [ ] User approves proceeding with apply

Step 5: Approval Gate

Present the analysis to the user and wait for explicit approval:

"Based on my analysis, this plan has [RISK LEVEL] risk. [Summary of key findings]. Do you want me to proceed with terraform apply? Please respond with 'approve' to continue."

NEVER proceed without explicit "approve" from user.

Step 6: Execute Apply (Only After Approval)

If and only if user explicitly approves:

terraform apply plan.out

Monitor output and report results.

Risk Categories

CRITICAL - Requires Extra Scrutiny

• Any resource destruction
• IAM policy changes
• Security group rule modifications
• Database modifications
• Encryption key changes
• Cross-account resource access

HIGH

• Network configuration changes
• Load balancer modifications
• Auto-scaling changes
• DNS record modifications

MEDIUM

• Instance type changes
• Tag modifications
• Non-critical configuration updates

LOW

• Pure additions with no dependencies
• Documentation-only changes

Common Patterns to Flag

1. Cascade Deletions: Resource deletion that triggers other deletions
2. State Drift: Plan shows changes that weren't in code
3. Dependency Chains: Changes that affect many downstream resources
4. Security Relaxation: Rules becoming more permissive
5. Cost Impact: Significant size/count changes

Memory Integration

Before analysis, query memory for:

• Similar changes in this project's history
• Known issues with affected resources
• Past incidents related to this type of change

After completion, store:

• Outcome of this change (success/failure)
• Any issues encountered
• User preferences learned

Verification Checklist

Before presenting to user, verify:

• AWS profile matches environment
• Plan was generated successfully
• All agents completed analysis
• Risk level is accurately assessed
• All destruction operations are highlighted
• Security implications are documented