Infrastructure development can mirror software development patterns: greenfield projects create new Terraform configurations for fresh AWS resources, while brownfield projects import existing infrastructure under Terraform management. Both scenarios require extensive context—provider details, resource dependencies, your security patterns, and import procedures. And the results can still feel like hallucinations when you run the next terraform validate or plan.

* The Context Bloat Challenge: AI agents need comprehensive knowledge to handle Terraform workflows effectively, but loading multiple MCP servers consumes 50,000+ tokens (40% of context window) before generating any code. This "context rot" degrades response quality and slows development velocity.

The Solution: Dynamic, workflow-driven context loading eliminates bloat while maintaining expertise with kiro skills and steering docs. Bundling MCP tools with curated steering documents that encode infrastructure best practices, ensuring agents have the right knowledge at the right time without overwhelming the context window.

Posts:
- https://manuchandrasekhar.com/posts/2026/02-06-adding-tf-skills/
- https://manuchandrasekhar.com/posts/2026/01-27-kiro-terraform-skills/