cost-lean-orchestration
降低主模型上下文成本，适合复杂工程会话。
在主模型外用子代理分担大量读取与搜索以节省Token。
在复杂工程任务里用“省 token”的模型路由：把海量文件读取、代码搜索、日志/文档扫描等交给 Sonnet 子代理去做，主模型只拿回紧凑的结构化发现用于综合判断与架构决策。适合要读很多文件、扫 repo、调大 bug、审大 diff 或多步骤工程工作时使用。特点是保护主模型的上下文窗口，把阅读成本交给廉价子代理，从而节省费用且聚焦高价值推理。
▸ 展开 SKILL.md 英文原文
Token-efficient model routing for complex engineering sessions. The main model (Opus / Fable, high reasoning) never bulk-reads or explores — it delegates all exploration, bulk file reading, codebase search, log scanning, dependency tracing, and documentation reading to Sonnet subagents that return compact structured findings, and it spends its own tokens only on synthesis, architecture, judgment, and final code. Use this skill in EVERY session that involves reading more than a couple of files, exploring or mapping a codebase, investigating a bug, auditing call sites, reviewing a large diff, digesting logs or docs, or any multi-step engineering task — even if the user never mentions cost, tokens, subagents, or delegation. This is a standing operating discipline, not an on-request feature.
帮我安装这个 skill：https://raw.githubusercontent.com/itzTiru/claude-code-skills-fable5/main/cost-lean-orchestration/SKILL.mdcurl -fsSL "https://raw.githubusercontent.com/itzTiru/claude-code-skills-fable5/main/cost-lean-orchestration/SKILL.md"# Cost-Lean Orchestration ## The core rule The context window of the main model is the most expensive real estate in the session. Every raw file, log dump, or search result loaded into it is paid for on every subsequent turn, and it crowds out the reasoning the main model is actually there for. So: **The main model synthesizes. Sonnet subagents read.** - Anything whose primary cost is *reading* (exploration, search, bulk file reading, log scanning, doc digestion, call-site auditing) → delegate to a **Sonnet** subagent. - Anything whose primary cost is *thinking* (architecture, tradeoffs, subtle correctness, root-cause judgment, final implementation of hard code, the answer to the