Anthropic's Prompt Audit Tool Flags Outdated Patterns in Claude Integrations, Cutting Costs 9% Further

Lance Martin, an engineer at Anthropic, announced /claude-api prompt-audit, a one-line Claude Code command that scans existing prompts, skills, and configuration files for dated instructions and proposes modernization fixes aligned with Opus 5.5. The tool is part of the open-source claude-api skill and audits CLAUDE.md, AGENTS.md, tool descriptions, and custom prompts for anti-patterns that degraded performance in earlier Claude generations but now waste tokens and reduce reliability on frontier models.