massimodeluisa/recursive-decomposition-skill

递归分解技能

在保持上下文较小的同时，分解长上下文任务。 基于递归语言模型研究的Claude代码、Codex、光标及兼容代理的代理技能

安装· 使用情况 · 工作原理· 评估 · 结构· 致谢

---

问题所在

庞大的代码库、数十个文档、冗长的报告：随着上下文的增长，模型会遗漏细节，通过猜测连接远处部分，导致准确性下降。递归语言模型论文称之为上下文腐烂。

它的作用

当任务是代码库范围的、多文档、一堆PDF或密集的聚合时，即使它符合窗口，技能也要求代理将输入视为查询环境，而非文本被吸收：

1. **读取任何东西前先调整输入大小。
2. 用搜索筛选搜索空间，而不是阅读。
3. 将剩余部分分块**成批5到10个文件或自然单元。
4. 递归在深度1，每批有一个子代理，每个子代理都有自包含的简报。子代理不生成子代理。
5. 验证合并后的答案，在一个小窗口中与来源对比。

Recursive Decomposition Skill

Decompose long-context tasks while keeping the context small. An Agent Skill for Claude Code, Codex, Cursor and compatible agents, based on the Recursive Language Models research

Install · Usage · How it works · Eval · Structure · Acknowledgments

---

The problem

Large codebases, dozens of documents, long reports: as the context grows, models miss details, link distant parts by guesswork and lose accuracy. The Recursive Language Models paper calls it context rot.

What it does

When a task is codebase-wide, multi-document, a pile of PDFs, or a dense aggregate, even if it fits the window, the skill makes the agent treat the input as an environment to query instead of text to swallow:

1. Size the input before reading anything.
2. Filter the search space with searches, not reads.
3. Chunk what remains into batches of 5 to 10 files or natural units.
4. Recurse at depth 1 with one sub-agent per batch, each with a self-contained brief. Sub-agents do not spawn sub-agents.
5. Verify the merged answer on a small window against the sources.
6. Synthesise programmatically, with file and line references.

PDFs go through anydoc first (npx -y @firecrawl/anydoc FILE -o .firecrawl/out.md), then grep the markdown. Cloud firecrawl parse is for OCR, -Q, or -S. A one-page convert job is not this skill.

Install

With the skills CLI:

npx skills add massimodeluisa/recursive-decomposition-skill