Eight GitHub Projects Form Coding-Agent Infrastructure
A commentary examined the top 20 projects on GitHub Trending and selected eight repositories it described as infrastructure for coding agents. Using a GitHub REST API snapshot dated November 19, 2026, the author reported about 1.35 million combined stars and grouped the projects into three layers: skills and methodology, orchestration, and harness, runtime or transport. The author cautioned that star counts are point-in-time popularity signals rather than evidence of usage or maturity, and separately discussed risks involving skill-package instructions, tool permissions, data transmission, skill conflicts and context growth. The selection criteria and three-layer framework are the author's interpretation, not an independently validated classification.
The report describes an emerging three-layer ecosystem around coding agents and highlights supply-chain, permission-boundary and context-management considerations. It recommends that teams review skill packages' required permissions and network activity and compare results with and without those skills before wider adoption.
昨天翻 GitHub 趋势榜,本来只是想找两个能用的小工具。
结果翻到一半发现一个挺反常的结构:榜单前列几乎看不到模型和应用,全都在解决同一个问题——怎么让写代码的 AI 代理把活干得更好。
20 个项目里,按"自身不产出业务价值、只为提升代理能力而存在"这条标准,我筛出 8 个。它们的 Star 加起来 约 135 万,且全部诞生于最近 14 个月。
这篇文章我想做三件事:把数据摆清楚、把 8 个项目归成三层抽象、说清楚为什么是现在,以及它带来的新风险。
一、方法与数据
不引用任何二手榜单。Star 数直接从 GitHub REST API 拉的,脚本如下:
#rea#superpowers#ECC#orca#t3code#dsh-Go#modern-web-guidance#Claude-BugHunter#Claude Code#Codex#Cursor CLI