Agent Skills › nduckmink/arkon › arkon-query

arkon-query

GitHub

基于 Arkon 知识库的查询技能，支持快速、标准和深度三种模式。通过搜索 Wiki 并仅在必要时深入原始来源，精准回答用户关于知识库的问题，提供结构化信息检索与综合。

skills/arkon-query/SKILL.md nduckmink/arkon

Trigger Scenarios

what do we know about find in KB look up query: arkon query search the wiki tell me about based on the KB

Install

npx skills add nduckmink/arkon --skill arkon-query -g -y

More Options

Use without installing

npx skills use nduckmink/arkon@arkon-query

指定 Agent (Claude Code)

npx skills add nduckmink/arkon --skill arkon-query -a claude-code -g -y

安装 repo 全部 skill

npx skills add nduckmink/arkon --all -g -y

预览 repo 内 skill

npx skills add nduckmink/arkon --list

SKILL.md

Frontmatter

{
    "name": "arkon-query",
    "description": "Answer questions using the Arkon knowledge base. Searches wiki first, drills into raw sources only when needed. Supports quick, standard, and deep modes. Triggers on: what do we know about, find in KB, look up, query:, arkon query, search the wiki, tell me about, based on the KB.",
    "allowed-tools": "mcp__arkon__search_wiki mcp__arkon__read_wiki_index mcp__arkon__read_wiki_page mcp__arkon__list_wiki_pages mcp__arkon__list_knowledge_types mcp__arkon__get_knowledge_type_docs mcp__arkon__list_sources mcp__arkon__get_source mcp__arkon__get_source_outline mcp__arkon__get_source_pages"
}

arkon-query: Query the Knowledge Base

The wiki has already done the synthesis work. Read strategically, answer precisely. Raw sources are a fallback — not the first stop.

Query Modes

Mode	Trigger	Tools used	Best for
Quick	`query quick: ...` or simple factual Q	`search_wiki` only	Direct lookups, names, dates
Standard	default	`search_wiki` → 2-4 pages	Most questions
Deep	`query deep: ...` or "thorough", "find everything"	Full wiki traverse + source drill-down	Cross-cutting synthesis

Quick Mode

Call search_wiki(query, top_k=5). If the top result has high similarity (≥ 80%) and its summary answers the question, respond immediately.
If not satisfied, call read_wiki_page(slug) for the top 1-2 results.
If still not found, say "Not in quick cache — run as standard query?"

Do not call read_wiki_index or source tools in quick mode.

Standard Query Workflow

search_wiki(query) — ranked list of pages. Read summaries.
read_wiki_page(slug) for the 2-4 most relevant results. Follow wikilinks [[slug]] at depth 1 only if they look essential.
Synthesize the answer in chat. Cite with page slugs: (Source: concept/onboarding).
If the question reveals a gap: "I don't have enough on X in the KB. Want to check the raw sources?"

Deep Mode

search_wiki + read_wiki_index to map coverage.
Read all relevant pages. Follow wikilinks at depth 2 if needed.
If wiki coverage is thin for a specific claim, drill into raw sources (see below).
Always cite both wiki pages and source documents used.

Source Drill-Down (citations or precise text)

Use when: the wiki has paraphrased something and you need the exact wording, or the user asks "what does the document actually say."

list_sources(knowledge_type=...)     → find source IDs by category
get_source(source_id)                → check metadata and page count
get_source_outline(source_id)        → find the right section (TOC)
get_source_pages(source_id, "5-7")   → read exact pages

Do not call get_source_pages without first checking get_source_outline on documents > 10 pages.

Browsing by Category

When the user asks "what's in the X category" or wants to explore a knowledge type:

list_knowledge_types()               → see all categories + doc counts
get_knowledge_type_docs(slug)        → docs in that category
list_wiki_pages(knowledge_type=slug) → wiki pages tagged to that category

Token Discipline

Start with	When to stop
`search_wiki`	If summary has the answer
2-4 wiki pages	Usually sufficient for standard queries
`get_source_outline`	Before reading pages of a long source
`get_source_pages`	Read only the specific pages needed

Never read more than 6 wiki pages for a standard query. For deep mode, cap at 15 unless the question explicitly demands full coverage.

RBAC Notes

Your MCP token determines which knowledge types you can access.
search_wiki, read_wiki_page, and list_wiki_pages automatically filter to your allowed types.
list_sources and get_source_pages enforce per-source scope.
If you get "Access denied" or "out of scope", the content exists but is outside your token's permissions — tell the user and suggest they contact an admin.