https://zhanwei.wang/en/tags/llm-agent/Recent content in Llm-Agent on Zhanwei WangHugoen-USFri, 17 Apr 2026 00:00:00 +0800https://zhanwei.wang/en/posts/skill-cost-optimization/Fri, 17 Apr 2026 00:00:00 +0800https://zhanwei.wang/en/posts/skill-cost-optimization/<blockquote> <p>This article is based on a week-long optimization effort across three production Skills — <code>prd-analysis</code>, <code>system-design</code>, and <code>autoforge</code> — covering the full loop from token-level measurement to actual code changes. All numbers come from real JSONL session files, with inflation factors corrected.</p> </blockquote> <h2 id="why-skill-cost-deserves-its-own-treatment">Why Skill Cost Deserves Its Own Treatment</h2> <p>Generic &ldquo;LLM cost reduction&rdquo; articles usually talk about context pruning, cache warmup, and model downgrading. These apply to Skills too — but the Skill execution environment has several <strong>structural differences</strong>:</p>