Agent Development Framework FAQs

Question 1

What is the benefit of declarative vs. imperative instructions?

Accepted Answer

Declarative instructions describe the goal and success criteria rather than specific tool commands, allowing the model to choose the best path and reason more effectively about the task.

Question 2

How do I fix memory crashes when running multiple Claude Code agents?

Accepted Answer

You should increase the Node.js heap limit by adding 'export NODE_OPTIONS="--max-old-space-size=16384"' to your .bashrc or .zshrc file to provide up to 16GB of memory.

Question 3

Why should I limit an agent's access to the Bash tool?

Accepted Answer

Limiting Bash access prevents 'approval fatigue.' Agents without Bash default to using the Write and Edit tools, which requires fewer manual user approvals per agent run.

Question 4

How do I make Claude auto-delegate to my custom agent?

Accepted Answer

Use the 'Strong Trigger Pattern' in your agent description, including specific phrases like 'MUST BE USED when' and 'Use PROACTIVELY for' alongside specific keywords to trigger automatic task handoff.

Question 5

Should I use Task sub-agents or remote API calls?

Accepted Answer

Always prefer Task sub-agents. Unlike remote API calls, sub-agents have full tool access, can read files iteratively, and can reason across your local project documents.

Agent Development Framework

Key Features

Use Cases

Agent Development Framework

Key Features

Use Cases