Anthropic cuts Claude Code prompt 80%, adds /doctor
Anthropic updated the Claude Code agent harness, reducing its default system prompt size by 80% in favor of progressive skill disclosure. The update introduces a `/doctor` command to help developers right-size context, eliminate over-constrained rules, and optimize prompt configuration files such as `CLAUDE.md`.
Trimming system prompt overhead is a major win for developer velocity and context window efficiency in agentic coding environments.
* Cutting prompt overhead by 80% preserves valuable context tokens for codebase reasoning and reduces output latency.
* Progressive skill disclosure enables agents to retrieve instructions dynamically only when needed rather than loading monolithic prompts upfront.
* The `/doctor` command provides actionable insights for developers to clean up legacy rules and maintain lightweight configuration files.
DISCOVERED
1h ago
2026-07-26
PUBLISHED
1h ago
2026-07-26
RELEVANCE
AUTHOR
Discover AI
