Claude Code – Everything You Can Configure That the Docs Don't Tell You
16 hours ago
- #AI Development Tools
- #Undocumented Features
- #Claude Code
- Claude Code's source code reveals numerous undocumented features that extend its capabilities beyond the official documentation.
- Hooks can return JSON on stdout to modify behavior in real time, such as rewriting commands, setting permissions, or injecting context.
- Additional hook fields include 'once', 'async', and 'asyncRewake', which control execution frequency and blocking behavior.
- Skills support undocumented frontmatter fields like 'model' for model overrides, 'effort' for reasoning depth, and 'hooks' for scoped automation.
- Agents can use 'memory' for persistent storage across sessions, 'color' for UI customization, and 'omitClaudeMd' to ignore project conventions.
- The auto-mode 'YOLO Classifier' can be configured with plain English descriptions in the 'environment' array to inform safety decisions.
- Self-improvement features like 'autoMemoryEnabled' and 'autoDreamEnabled' create a learning loop by consolidating session memories.
- Magic Docs can be auto-maintained with specific scoping instructions, and permission rules support glob patterns for fine-grained control.
- The 'context: fork' setting allows background execution of heavy tasks, but model consistency is key for prompt caching efficiency.