Aviator Runbooks is a plan-first AI coding tool that uses Claude to generate a structured spec before writing any code. Instead of jumping straight into implementation, users provide a prompt, review an editable plan, then let the agent execute step-by-step and open a PR. This approach targets brownfield and legacy codebases where unguided AI agents tend to make undocumented micro-decisions and produce inconsistent output. Key features include one-shot mode for smaller tasks, automatic build/test failure recovery, GitHub integration, and support for custom Claude API keys or AWS Bedrock.
Table of contents
The Issue of Prompting Without a DirectionRunbooks Workflow: Plan → Execute → CollaborateA Real World ExampleNext Steps and ResourcesFrequently Asked Questions (FAQ)Sort: