Agent Skills › samhjn/Palvia › Deep Research

Deep Research

GitHub

提供系统化深度研究方法论，指导通过浏览器和脚本工具搜索、评估来源并综合信息，生成结构化研究报告。

Palvia/Resources/BuiltInSkills/deep-research/SKILL.md samhjn/Palvia

Trigger Scenarios

需要深入调研特定主题要求多源验证与综合分析报告

Install

npx skills add samhjn/Palvia --skill Deep Research -g -y

More Options

Non-standard path

npx skills add https://github.com/samhjn/Palvia/tree/develop/Palvia/Resources/BuiltInSkills/deep-research -g -y

Use without installing

npx skills use samhjn/Palvia@Deep Research

指定 Agent (Claude Code)

npx skills add samhjn/Palvia --skill Deep Research -a claude-code -g -y

安装 repo 全部 skill

npx skills add samhjn/Palvia --all -g -y

预览 repo 内 skill

npx skills add samhjn/Palvia --list

SKILL.md

Frontmatter

{
    "name": "Deep Research",
    "palvia": {
        "tags": [
            "research",
            "analysis",
            "methodology"
        ],
        "version": "1.0"
    },
    "description": "Systematic research methodology with source evaluation and synthesis"
}

Deep Research Skill

When asked to research a topic, follow this methodology:

Research Tools (in order of preference)

Browser tools (primary — most reliable for web research):
- browser_navigate(url:) → open a URL in the in-app browser
- browser_get_page_info(include_content: true) → read page text, links, and interactive elements
- browser_extract(selector:) → extract specific elements (headlines, article bodies, etc.)
- Search strategy: navigate to https://www.google.com/search?q=YOUR+QUERY to find sources, then visit promising results
fetch_and_extract (quick plain-text fetch — may fail due to sandbox network restrictions; if it returns HTTP 0 or network errors, switch to browser tools immediately)
Post-processing scripts (use run_snippet to execute):
- extract_links — parse HTML to extract follow-up URLs for deeper investigation
- summarize_text — condense long text into key points ranked by importance

Process

Decompose the question into 2–5 focused sub-questions
Search — start with a search engine query via browser, then visit the most promising results
Gather — read each source page with browser_get_page_info; use browser_extract for specific content; follow links for deeper investigation
Evaluate — assess source credibility (official docs > research papers > established news > blogs), recency, and potential biases
Synthesize — organize findings into a coherent analysis, cross-referencing multiple sources
Cite — reference every key claim to its source URL with a confidence level (High / Medium / Low)

Iterative Deepening

After the first pass, identify gaps or conflicting information
Search for additional sources to fill gaps or resolve conflicts
Stop when key claims have 2–3 corroborating sources, or when additional sources add no new information

Output Format

Executive Summary — 2–3 sentence overview of key findings
Findings — organized by sub-topic, each with inline source citations
Source Table — list of sources used with credibility rating (High / Medium / Low)
Uncertainties — flag conflicting information and knowledge gaps
Conclusions — actionable takeaways

Guidelines

Prefer primary sources (official docs, research papers, raw data) over secondary (news articles, blog posts)
Note when information might be outdated — check publication dates
Distinguish clearly between established facts, expert opinions, and speculation
When a tool fails, switch to an alternative immediately — do not retry the same failing tool
Save key findings to MEMORY.md for future reference