<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Token Miser</title><description>Practical LLM cost optimization for builders who are tired of watching their API bill climb.</description><link>https://tokenmiser.uk/</link><item><title>The Complete Guide to Claude Code Agent Optimization</title><link>https://tokenmiser.uk/blog/claude-code-agent-optimization/</link><guid isPermaLink="true">https://tokenmiser.uk/blog/claude-code-agent-optimization/</guid><description>How Claude caches work at the byte level, how to structure agents and rules for maximum efficiency, and practical techniques that saved us 60-87% on agent token costs across 9 production repos.</description><pubDate>Thu, 09 Apr 2026 00:00:00 GMT</pubDate></item><item><title>Skills vs Rules vs Reads: The Claude Code Subagent Caching Cheatsheet</title><link>https://tokenmiser.uk/blog/claude-code-skills-vs-rules-subagent-cheatsheet/</link><guid isPermaLink="true">https://tokenmiser.uk/blog/claude-code-skills-vs-rules-subagent-cheatsheet/</guid><description>Where you put shared content in Claude Code determines whether subagents see it at all — and which cache tier it lands in. Here is the decision tree nobody wrote down.</description><pubDate>Tue, 14 Apr 2026 00:00:00 GMT</pubDate></item><item><title>Lint Your CLAUDE.md: Stop Paying for Stale Context Every Session</title><link>https://tokenmiser.uk/blog/lint-your-claude-md/</link><guid isPermaLink="true">https://tokenmiser.uk/blog/lint-your-claude-md/</guid><description>Your CLAUDE.md is probably billing you for files that no longer exist. Here is how to audit it in 10 minutes and save hundreds of thousands of tokens per month.</description><pubDate>Tue, 14 Apr 2026 00:00:00 GMT</pubDate></item><item><title>Use Local Models as a Compression Layer to Cut Your Claude Bill</title><link>https://tokenmiser.uk/blog/local-models-as-compression-layer/</link><guid isPermaLink="true">https://tokenmiser.uk/blog/local-models-as-compression-layer/</guid><description>Stop paying Claude $5/MTok to read a 528 KB GitHub issue. Route big content through a cheap local model first, hand Claude the summary, save 95%+ on tokens.</description><pubDate>Tue, 14 Apr 2026 00:00:00 GMT</pubDate></item><item><title>The MCP Tool Prefix Trap: How Inherited Tools Silently Break Your Cache</title><link>https://tokenmiser.uk/blog/mcp-server-tool-prefix-cache-trap/</link><guid isPermaLink="true">https://tokenmiser.uk/blog/mcp-server-tool-prefix-cache-trap/</guid><description>Every subagent spawn ships 7,000+ tokens of MCP tool definitions you never use. One frontmatter line fixes it — and unlocks cross-agent cache sharing.</description><pubDate>Tue, 14 Apr 2026 00:00:00 GMT</pubDate></item><item><title>Prompt Caching Basics: Stop Paying Full Price for the Same Tokens Twice</title><link>https://tokenmiser.uk/blog/prompt-caching-basics/</link><guid isPermaLink="true">https://tokenmiser.uk/blog/prompt-caching-basics/</guid><description>Every API call re-sends your entire prompt by default. Prompt caching fixes that at 10% of normal input cost. Here is how it works, what breaks it, and how to verify it.</description><pubDate>Tue, 14 Apr 2026 00:00:00 GMT</pubDate></item><item><title>Track Your Claude Code Spend with OTel + Jaeger (and the Gotcha That Wastes Your Morning)</title><link>https://tokenmiser.uk/blog/track-claude-code-spend-otel-jaeger/</link><guid isPermaLink="true">https://tokenmiser.uk/blog/track-claude-code-spend-otel-jaeger/</guid><description>Claude Code ships OTel support. Most tooling silently returns zero tokens. Here is the full setup, the classifier bug that causes it, and the three queries worth your time.</description><pubDate>Tue, 14 Apr 2026 00:00:00 GMT</pubDate></item></channel></rss>