跳到主要内容
    ↑↓ 选择↵ 打开esc 关闭
    中文English
    zhanjian0217

    Skill Search

    v0.1.4工具与命令
    opencode-skill-search

    BM25-powered skill search plugin for OpenCode — discover and load agent skills by keyword

    GitHub 星标

    0

    月装机量

    66

    近 7 天 9

    综合评分SCORE

    23.5

    生态多维模型

    最近提交

    4 个月前

    2026-03-27

    快速安装与配置

    opencode.json

    写入当前项目的 opencode.json,只对这个仓库生效。

    opencode.json

    {
      "$schema": "https://opencode.ai/config.json",
      "plugin": ["opencode-skill-search@0.1.4"]
    }

    opencode 启动时会通过内嵌运行时自动加载 npm 依赖并缓存至本地目录,无需手动在全局环境执行安装。

    繁體中文

    opencode-skill-search

    BM25-powered skill search plugin for OpenCode — discover and load agent skills by keyword.

    Why?

    OpenCode's built-in skill tool embeds the full list of available skills in its tool description, consuming tokens on every request. This plugin replaces that with an on-demand skill_search tool that uses BM25 ranking to find relevant skills only when needed.

    What it does

    • skill_search tool — search installed skills by keyword with BM25 relevance ranking
    • System prompt injection — reminds the agent to search for skills before tackling non-trivial tasks
    • Skill tool trimming — replaces the verbose skill list in the skill tool description with a pointer to skill_search

    Install

    Add to your opencode.json:

    {
      "plugin": ["opencode-skill-search"]
    }
    

    OpenCode will auto-install the package on next launch.

    Scanned directories

    The plugin looks for SKILL.md files in these directories:

    Directory Source
    <project>/.opencode/skills/ Project OpenCode
    ~/.config/opencode/skills/ Global OpenCode
    <project>/.claude/skills/ Project Claude-compatible
    ~/.claude/skills/ Global Claude-compatible
    <project>/.agents/skills/ Project Agent-compatible
    ~/.agents/skills/ Global Agent-compatible
    ~/.cache/opencode/node_modules/superpowers/skills/ Installed skill packs

    Usage

    Once installed, the agent will automatically use skill_search before tasks. You can also call it directly:

    skill_search({query: "testing"})
    skill_search({query: "docker deployment", max_results: 3})
    

    License

    MIT