菜单
Medium-depth literature search — read AI-summarized reports for every paper analyzed
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/yogsoth-ai/de-anthropocentric-research-engine --skill literature-search命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Strategy: Dialectic engine retuned for truth-seeking, not survival. A defender steelmans a claim into its MOST falsifiable form, a critic attacks to refute it, a judge classifies the exchange into BROKEN/CORROBORATED/UNFALSIFIABLE — the judge does NOT pick a winner or score persuasiveness. Methods: Irving debate (repurposed), Toulmin argumentation, Mayo severe testing.
Campaign: Logical extreme and boundary testing via reductio ad absurdum and edge-case analysis. Core question: Does this artifact collapse under logical limits and boundary conditions? Methods: Lakatos 1976, Dutilh Novaes 2016, BVA, Flyvbjerg Critical Case, Popper.
Campaign for mapping argument structures — extract claims, link evidence, assess strength, synthesize positions. Produces argument graphs in the wiki vault.
基于 SOC 职业分类
正在显示 SKILL.md
| name | literature-search |
| description | Medium-depth literature search — read AI-summarized reports for every paper analyzed |
| type | sop |
| layer | sop |
| agents | ["alphaxiv","semantic-scholar"] |
| tools | {"alphaxiv":["discover_papers","get_paper_content"],"semantic-scholar":["relevanceSearch","paper","paperBatch","citations","references"]} |
| input | query (string), scope (survey | gap-analysis | background) |
| output | PaperAnalysis[] with metadata + AI summary content |
| dependencies | {"sops":["literature-research"]} |
Medium-depth reading. Understand methods, contributions, and findings via AI-generated summary reports. Suitable for literature surveys, gap analysis, and building background knowledge.
Use this when you need to:
This skill reads AI-summarized reports — not raw full text. For rigorous analysis requiring raw text, use literature-research.
| Tool | Purpose | Returns |
|---|---|---|
alphaxiv.discover_papers | Primary search — arXiv semantic search | Ranked paper list with metadata |
ss.relevanceSearch | Supplementary search — non-arXiv papers | Title, abstract, authors, citationCount |
ss.paper / ss.paperBatch | Metadata enrichment | Citation count, DOI, S2 ID, externalIds |
ss.citations | Papers that cite this paper (incoming) | Citing paper list with context |
ss.references | Papers this paper cites (outgoing) | Referenced paper list |
alphaxiv.get_paper_content | AI summary report (fullText: false) | Structured AI-generated paper report |
For EVERY paper selected for analysis, you MUST call:
alphaxiv.get_paper_content(url: arxiv_url, fullText: false)
This returns an AI-generated summary report optimized for LLM consumption.
PROHIBITED:
REQUIRED:
Primary (arXiv):
alphaxiv.discover_papers(
keywords: ["keyword1", "keyword2", "keyword3"],
question: "Detailed description of papers needed",
difficulty: 5
)
Supplementary (non-arXiv):
ss.relevanceSearch(
query: "search terms",
limit: 20,
year: "2022-2024"
)
For papers found via alphaxiv, enrich with citation data:
ss.paperBatch(
paper_ids: ["ARXIV:2301.xxxxx", "ARXIV:2302.xxxxx", ...]
)
Returns: citationCount, DOI, S2 ID for each paper.
Choose top N papers (typically 5-15) based on:
For each selected paper:
alphaxiv.get_paper_content(
url: "https://arxiv.org/abs/XXXX.XXXXX",
fullText: false
)
fullText: false (default) returns an AI-generated intermediate report:
To find related work not caught by keyword search:
ss.citations(paper_id: "ARXIV:XXXX.XXXXX", limit: 50)
ss.references(paper_id: "ARXIV:XXXX.XXXXX", limit: 50)
Filter results by year and citation count, then repeat Steps 3-4 for promising papers.
fullText: false (default) — AI-generated report, faster, structuredfullText: true — raw extracted text, slower, complete (use in literature-research, not here)https://arxiv.org/abs/XXXX.XXXXX), PDF URL, alphaXiv URL2301.12345 → ARXIV:2301.12345)offset and limit for pagination# Step 1: Search
alphaxiv.discover_papers(
keywords: ["vision transformer", "attention", "ViT"],
question: "Papers proposing or analyzing attention mechanisms in vision transformers",
difficulty: 5
)
ss.relevanceSearch(query: "vision transformer attention mechanism", limit: 15, year: "2022-2024")
# Step 2: Enrich
ss.paperBatch(paper_ids: ["ARXIV:2010.11929", "ARXIV:2103.14030", ...])
# Step 3: Select top 8 by citation count + relevance
# Step 4: Read each
alphaxiv.get_paper_content(url: "https://arxiv.org/abs/2010.11929") # ViT
alphaxiv.get_paper_content(url: "https://arxiv.org/abs/2103.14030") # Swin
# ... repeat for all 8
# Step 5: Expand via citations of ViT
ss.citations(paper_id: "ARXIV:2010.11929", limit: 30)
# Step 1: Broad search
alphaxiv.discover_papers(
keywords: ["LLM", "efficient inference", "quantization", "pruning"],
question: "Methods for making large language model inference faster or cheaper",
difficulty: 6
)
# Step 2-4: Enrich, select 10, read AI summaries
# Step 5: Check what recent papers cite the seminal works
ss.citations(paper_id: "ARXIV:2210.17323", limit: 50) # GPTQ
ss.citations(paper_id: "ARXIV:2306.00978", limit: 50) # AWQ
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use |
|---|---|
| literature-research | Deep literature research — raw full text reading and targeted PDF queries for rigorous analysis |