Agent Skills › davidliuk/graph-of-skills › vector-skills-retriever

vector-skills-retriever

GitHub

通过向量相似度检索本地技能工作区中的相关技能包，支持子问题细化查询。用于在缺乏上下文时获取专业领域知识、脚本或参考信息，辅助后续开发任务。

skills/vector-skills-retriever/SKILL.md davidliuk/graph-of-skills

Trigger Scenarios

需要特定领域专业知识但当前上下文不足需要查找并引用本地存储的脚本或参考资料遇到复杂问题需先检索相关技能再执行

Install

npx skills add davidliuk/graph-of-skills --skill vector-skills-retriever -g -y

More Options

Use without installing

npx skills use davidliuk/graph-of-skills@vector-skills-retriever

指定 Agent (Claude Code)

npx skills add davidliuk/graph-of-skills --skill vector-skills-retriever -a claude-code -g -y

安装 repo 全部 skill

npx skills add davidliuk/graph-of-skills --all -g -y

预览 repo 内 skill

npx skills add davidliuk/graph-of-skills --list

SKILL.md

Frontmatter

{
    "name": "vector-skills-retriever",
    "inputs": [
        "task description or subproblem summary"
    ],
    "outputs": [
        "a ranked skill bundle with full instructions and source paths under \/opt\/graphskills\/skills"
    ],
    "description": "Retrieve a bounded bundle of relevant external skills from the local skill workspace using vector embedding similarity only.",
    "allowed-tools": [
        "shell",
        "python3"
    ],
    "compatibility": [
        "claude-code",
        "codex",
        "gemini-cli"
    ]
}

Purpose

Use this skill when the task likely needs specialized domain knowledge, scripts, or references that are not already obvious from the current context.

This retriever uses embedding similarity only. It does not use graph edges, graph propagation, or lexical expansion.

Retrieve Relevant Skills

Run:

OPENROUTER_API_KEY="${OPENROUTER_API_KEY:-$OPENAI_API_KEY}" \
OPENAI_API_KEY="${OPENROUTER_API_KEY:-$OPENAI_API_KEY}" \
OPENAI_BASE_URL=https://openrouter.ai/api/v1 \
GOS_EMBEDDING_MODEL=openai/text-embedding-3-large \
GOS_EMBEDDING_DIM=3072 \
vectorskills-query "short description of the task or current subproblem"

Diagnostic rule: do not suppress stderr and do not replace failures with fallback text. If vectorskills-query fails, keep the original stderr visible and note the real exit code so the failure remains diagnosable.

Useful flags:

vectorskills-query "debug spring boot jakarta migration build errors" --top-n 5 --max-context-chars 9000
vectorskills-query "extract text from receipts into xlsx" --json

How To Use The Results

Start with a short task-level query.
Read the returned skill bundle.
Follow the retrieved skill instructions and inspect any referenced files under /opt/graphskills/skills/<skill-name>/.
Re-query with a narrower subproblem if needed.