I'm new to using Claude Code at my workplace. While it handles coding tasks well about 85% of the time, I notice it struggles with more complex issues that require a deeper understanding of our organization's infrastructure and setups. I've attempted various tools and open-source solutions, but they haven't provided the detailed organizational knowledge I need. Is there anyone with experience in working with coding agents who can share effective strategies or solutions for this problem?
5 Answers
Consider allowing your agent to read your manifests and deployment scripts while also having it ask questions in plan mode. You might be surprised by how quickly it can get you to a decent solution.
One approach is to use Agent Skills to document the essential organizational knowledge and feed it into the LLM's context using a technique known as progressive disclosure.
I haven't faced many challenges using my copilot chatbot for these conversations. It's occasionally off, but I can easily switch between different LLMs as needed. Plus, I have a solid understanding of my environment since I work at a big company.
If you're working within an enterprise framework, you should let the agent review your code repositories, development documentation, and relevant policies, especially security protocols. I've had success using prompts like: 'Search our entire GitLab instance and see how most teammates have solved this issue.' It doesn’t always give the answer but often gets me on the right track.
It's pretty simple! Just have it read your Infrastructure as Code (IaC). LLMs can interpret configurations from tools like Terraform and Ansible really well. Kubernetes can be tricky since the live setup might differ from the declared configurations, so I try to keep everything as declarative as possible, but it still works.
What do you mean when you say k8s manifests aren't as declarative as you'd like?