Anthropic added /claude-api prompt-audit to Claude Code

Anthropic added the /claude-api prompt-audit command to the official claude-api skill for Claude Code, which scans CLAUDE.md, skills, agent.md, tool descriptions, and prompts for "antipatterns" from older models and suggests removing them.

Compiled by Sergey KostenchukPublished 2026-09-23Updated 2026-09-24

2026-09-23 Coding Anthropic

🛠 Anthropic added /claude-api prompt-audit to Claude Code

The command in the official claude-api skill scans CLAUDE.md, skills, agent.md, and prompts for "antipatterns" from older models — verification rituals, all-caps intensifiers ("CRITICAL: YOU MUST"), forced scratchpad reasoning, conflicting rules, and outdated thinking settings — and suggests removing them.

🌍 Instructions that "patched" the weaknesses of older models are harmful on frontier models: they waste tokens, break prompt caching, and conflict with built-in reasoning, so auditing harness configurations becomes a step in migrating to a new model.

👤 Run it in your repositories: on the Opus 4.8 → Opus 5.5 migration benchmark, switching models yielded ~18% savings, while prompt-audit added another ~9% savings and ~2 percentage points of accuracy.

Source 1: https://claude.com/blog/reducing-cost-and-improving-performance-with-claude-platform Source 2: https://github.com/anthropics/skills/tree/main/skills/claude-api

Sources