A Claude Code skill was eating 200,000 tokens before answering a single question
A Claude Code skill designed to help developers work with Anthropic’s API was consuming more than 200,000 tokens to load. The post A Claude Code skill was eating 200,000 tokens before answering a single question appeared first on The New Stack .
An Anthropic skill designed to assist developers with Claude's API was consuming over 200,000 tokens to load a single question, according to a July 7 GitHub issue. The issue reported that the skill embedded reference files and language documentation directly into its body, totaling roughly 120,000 tokens. One migration document accounted for an estimated 36,000 tokens.
Despite similar reports, the issue was marked "not planned." A subsequent bug report in August highlighted an expensive fallback when the skill couldn't detect a project language, loading documentation for multiple languages and shared Markdown files, totaling 812,650 bytes. The bundle contained only one needed file of 32,954 bytes.
Inlining the large reference library forced the agent to read all 812 KB for every request, even when most was irrelevant. This practice creates a hidden overhead problem, as the skill's content becomes a fixed cost that developers may not see. Cutting the skill to roughly 25,000 tokens helps alleviate the issue. Anthropic's skill documentation suggests dividing loading into three stages: small metadata, SKILL.md body when triggered, and bundled resources pulled in during the task.
Written by urgent.news from The New Stack's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
Also reported by 1 other outlet
- Claude Code weekly limits reduce by a third tomorrow support.claude.com
