菜单
Use when delegating GitHub repository reconnaissance to the ghx sidecar service: ask repo questions in plain English and receive compact evidence reports.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/gkoreli/ghx --skill ghx-recon命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
| name | ghx-recon |
| description | Use when delegating GitHub repository reconnaissance to the ghx sidecar service: ask repo questions in plain English and receive compact evidence reports. |
| version | 1.0.0 |
| author | ghx contributors |
| license | MIT |
| metadata | {"hermes":{"tags":["github","code-exploration","sidecar","repository-recon"],"related_skills":[]}} |
ghx is a specialist agent that explores GitHub for you: ask a repo question in plain English and it returns a compact evidence report. Never explore a remote repo step-by-step yourself — delegate the whole question and let ghx do the reading, mapping, and searching under the hood.
Two entry points, same service:
recon tool — recon(question, repo?).ghx sidecar ask "<question>".State the goal and name the repo when you know it — "How does hono implement middleware chaining, and which files define it?" beats "run grep for middleware". Omit the repo for discovery questions — "which repos or libraries do X" — and ghx sweeps GitHub, verifies the top candidates by reading them, and ranks them. One investigation per question; follow up rather than bundling.
Every answer ends with an artifacts pointer (session dir + trace id) — the on-disk audit trail behind the report. Follow-ups stay on the right investigation automatically; you never manage sessions. A fresh question takes tens of seconds; follow-ups are faster.