一键导入
reference-researcher
Advanced research strategies, import patterns, crawl configuration, and staleness management for reference topics.
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。
菜单
Advanced research strategies, import patterns, crawl configuration, and staleness management for reference topics.
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。
基于 SOC 职业分类
Create and improve file-based CRON jobs for a GHOST in T-KOMA. Use when you need to add or update `workspace/cron/*.md` jobs with schedule, pre-model tools, and a clear objective.
Create and manage structured knowledge notes, diary entries, and identity files. Use when you need to persist important information, curate inbox captures, or maintain a knowledge base.
Guide for creating effective Agent Skills for t-koma. Use when the user wants to create a new skill or update an existing skill.
| name | reference-researcher |
| description | Advanced research strategies, import patterns, crawl configuration, and staleness management for reference topics. |
| license | MIT |
| metadata | {"author":"t-koma","version":"2.0"} |
You are a research specialist. This skill covers advanced strategies for building high-quality reference topics. Basic reference_write and reference_import usage is in the system prompt — this skill adds depth.
Follow this priority order to understand a topic before importing:
Find documentation first: Use web_search to find official docs. Many projects
have a dedicated docsite repo (e.g. org/docs, org/website) that contains more
useful content than the code repo itself.
web_search(query="dioxus official documentation site")
Read key doc pages: Use web_fetch to read specific pages.
web_fetch(url="https://dioxuslabs.com/learn/0.6/", prompt="Key concepts and getting started?")
Find the code repository: Use gh search repos to locate the main repo.
run_shell_command(command="gh search repos 'dioxus' --language=Rust --sort=stars --limit=5")
Read repo metadata: Use gh api for description, stars, topics.
run_shell_command(command="gh api repos/DioxusLabs/dioxus --jq '.description, .stargazers_count, .topics'")
Understand before creating: Read enough to write a meaningful topic description and pick the right sources.
The embedding system handles large codebases well (tree-sitter chunking, hybrid search). Default to full repo imports:
{
"title": "Dioxus - Rust UI Framework",
"body": "Dioxus is a portable, performant framework for building cross-platform UIs...",
"sources": [
{
"type": "git",
"url": "https://github.com/DioxusLabs/dioxus",
"ref": "main"
},
{
"type": "git",
"url": "https://github.com/DioxusLabs/docsite",
"ref": "main",
"role": "docs"
},
{ "type": "web", "url": "https://dioxuslabs.com/learn/0.6/" }
],
"tags": ["rust", "ui", "framework", "dioxus"],
"max_age_days": 30
}
docs: Documentation content. Boosted 1.5x in search.code: Source code. Normal ranking.web → docs, git → code."role": "docs"
explicitly.Retry with a paths filter. Prioritize:
README.md — project overviewdocs/ — official documentationexamples/ — practical usage patterns{
"sources": [
{
"type": "git",
"url": "https://github.com/DioxusLabs/dioxus",
"ref": "main",
"paths": ["README.md", "docs/", "examples/"]
}
]
}
Combine git repos, web pages, and crawls in a single import. Look for separate documentation repos — they often have better content than the main repo's docs:
{
"sources": [
{ "type": "git", "url": "https://github.com/org/library", "ref": "main" },
{
"type": "git",
"url": "https://github.com/org/library-docs",
"ref": "main",
"role": "docs"
},
{
"type": "crawl",
"url": "https://library.dev/docs/",
"max_depth": 2,
"max_pages": 50
}
]
}
Use "type": "crawl" to automatically discover and fetch linked pages from a
documentation site:
max_depth (default 1, max 3): How many link-hops from the seed URL.max_pages (default 20, max 100): Maximum pages to fetch.When to crawl vs list individual pages:
The body is passed IN FULL to the LLM as context. Write it as a concise briefing:
Do NOT include code examples in the body — those belong in reference files.
Reference topics are shared notes. Create the topic note first with note_write, then
import sources:
note_write(action="create", scope="shared", title="Dioxus", body="Description...")reference_import(title="Dioxus", sources=[...])The reference_import tool creates the topic note automatically. For manual workflows
(e.g. reference_write), the topic note must exist first.
To update a topic's description or tags, use note_write(action="update").
Use reference_manage with action update to manage file status:
problematic when you find partial inaccuracies.obsolete when they would actively mislead.note_write(action="update").reference_manage(action="update", note_id="abc123",
status="problematic", reason="API examples use v0.4 syntax, current is v0.6")