Find a page by symptom¶

Search the platform by the problem you are trying to solve instead of by the concept that solves it. Each entry below is a real question a learner has asked or a real symptom an engineer has hit; click through to the page that addresses it directly.

Can I just put rules in the system prompt¶

• Prompt-injection defenses

Do I need LangChain to build an agent¶

• Build a tool-use loop without a framework

Do I need a vector database or is Postgres enough¶

• Production-grade RAG with hybrid retrieval + reranking + eval

How do I cap agent cost and turns¶

• Build a tool-use loop without a framework

How do I connect one client to multiple MCP servers¶

• Build an MCP client + tool-use loop

How do I debug a slow or expensive AI agent run¶

• Agent observability with OpenTelemetry

How do I defend my agent against adversarial users¶

• Prompt-injection defenses

How do I detect and heal config drift¶

• Build a GitOps sync loop

How do I enforce a Pod policy before objects land in etcd¶

• Build an admission webhook

How do I extend the Kubernetes API with my own resources¶

• Build an operator with a Custom Resource Definition

How do I inject defaults into Pod specs cluster-wide¶

• Build an admission webhook

How do I keep my SKILL.md from blowing up the context budget¶

• Build a Claude Skill from scratch

How do I let Claude or GPT call my code¶

• Build an MCP server from scratch

How do I make a workflow that Claude will follow consistently¶

• Build a Claude Skill from scratch

How do I make sure my agent never emails the wrong person¶

• Prompt-injection defenses

How do I make the model literally unable to break my schema¶

• Structured output with grammar-constrained decoding

How do I measure RAG quality¶

• Production-grade RAG with hybrid retrieval + reranking + eval

How do I parallelize independent agent work¶

• Multi-agent supervisor pattern

How do I parallelize tool calls in an agent loop¶

• Build a tool-use loop without a framework

How do I parse Anthropic SSE events by hand¶

• Streaming agent with mid-stream tool use

How do I run a custom scheduler alongside the default one¶

• Build a custom scheduler

How do I trace LLM calls across services¶

• Agent observability with OpenTelemetry

How do I write Claude Desktop or Cursor myself¶

• Build an MCP client + tool-use loop

How does Argo CD or Flux work under the hood¶

• Build a GitOps sync loop

How does Cilium enforce policies with eBPF¶

• Week 14: a NetworkPolicy → what eBPF actually does

How does Claude Code or ChatGPT stream tokens to the browser¶

• Streaming agent with mid-stream tool use

How does HPA decide the replica count¶

• Build an autoscaler from scratch

How does OpenAI strict mode actually work¶

• Structured output with grammar-constrained decoding

How does the Anthropic / OpenAI Agent SDK actually work under the hood¶

• Multi-agent supervisor pattern

How does the apiserver fit between everything else and etcd¶

• Bootstrap a Kubernetes control plane by hand

How does the scheduler actually pick a node¶

• Build a custom scheduler

How does the tool_use / tool_result round-trip work in the Anthropic API¶

• Build an MCP client + tool-use loop

How is a Kubernetes cluster actually built¶

• Bootstrap a Kubernetes control plane by hand

How is an MCP server different from a regular HTTP API¶

• Build an MCP server from scratch

How much does reranking actually help¶

• Production-grade RAG with hybrid retrieval + reranking + eval

I don't understand what a controller actually does¶

• Build a Kubernetes controller from scratch

Langfuse vs Phoenix vs Helicone vs Datadog for AI traces¶

• Agent observability with OpenTelemetry

Manual kubectl changes keep getting reverted¶

• Build a GitOps sync loop

My HPA is flapping¶

• Build an autoscaler from scratch

My LLM keeps emitting invalid JSON¶

• Structured output with grammar-constrained decoding

My RAG bot gives wrong answers¶

• Production-grade RAG with hybrid retrieval + reranking + eval

My agent keeps losing context across stages¶

• Multi-agent supervisor pattern

My process keeps getting OOMKilled¶

• Watch the OOM killer fire

My webhook is wedging the cluster¶

• Build an admission webhook

Pod stuck Pending forever¶

• Build a custom scheduler

Pods cannot reach each other¶

• Build pod networking from scratch (a minimal CNI)

Should I package this as a Skill or an MCP server¶

• Build a Claude Skill from scratch

What does a CNI plugin actually do¶

• Build pod networking from scratch (a minimal CNI)

What does a NetworkPolicy actually do at the kernel level¶

• Week 14: a NetworkPolicy → what eBPF actually does

What does input_json_delta mean and how do I dispatch tools mid-stream¶

• Streaming agent with mid-stream tool use

What is MCP and why does everyone keep mentioning it¶

• Build an MCP server from scratch

What is ReAct / plan-and-execute / reflection - are they real or just prompts¶

• Build a tool-use loop without a framework

What is a CRD and how do I write one¶

• Build an operator with a Custom Resource Definition

What is a Claude Skill¶

• Build a Claude Skill from scratch

What is an AI agent actually doing under the hood¶

• Build an MCP client + tool-use loop

What is an AI agent really doing under the hood¶

• Build a tool-use loop without a framework

What is gen_ai semantic convention in OpenTelemetry¶

• Agent observability with OpenTelemetry

What is grammar-constrained decoding¶

• Structured output with grammar-constrained decoding

What is indirect prompt injection and why is it worse than direct¶

• Prompt-injection defenses

What is prompt injection¶

• Prompt-injection defenses

What is the OOM killer and how does it pick a victim¶

• Watch the OOM killer fire

What is the reconcile loop¶

• Build a Kubernetes controller from scratch

What lives in etcd¶

• Bootstrap a Kubernetes control plane by hand

When should I use MCP versus calling the LLM API with my own tools¶

• Build an MCP server from scratch

When should I use Pydantic + retry vs constrained decoding¶

• Structured output with grammar-constrained decoding

When should I use multi-agent vs single-agent¶

• Multi-agent supervisor pattern

Why are my dependent resources not being cleaned up when I delete the parent¶

• Build an operator with a Custom Resource Definition

Why did my pod restart with reason OOMKilled¶

• Watch the OOM killer fire

Why does deleting a Deployment's Pod immediately recreate it¶

• Build a Kubernetes controller from scratch

Why does my AI feature feel laggy¶

• Streaming agent with mid-stream tool use

Why does the autoscaler scale up fast and down slowly¶

• Build an autoscaler from scratch

Why is etcd the only thing I need to back up¶

• Bootstrap a Kubernetes control plane by hand

Why is my pod stuck in ContainerCreating with a network error¶

• Build pod networking from scratch (a minimal CNI)

Why is my vector search missing exact-name queries¶

• Production-grade RAG with hybrid retrieval + reranking + eval