AGENT SKILL

deepxiv-baseline-table

Build a markdown baseline table for a research topic using deepxiv search, brief, head, and experiment-section reads, extracting paper title, URL, open-source status, datasets, benchmark scores, and other comparison-ready details.

License 许可
MIT
最近更新
2026-11-14
Stars
★ 3
Forks
0

01 前言 Front Matter

name
deepxiv-baseline-table
description
Build a markdown baseline table for a research topic using deepxiv search, brief, head, and experiment-section reads, extracting paper title, URL, open-source status, datasets, benchmark scores, and other comparison-ready details.

03 SKILL.md 正文

DeepXiv Baseline Table

Use this skill when the user wants to map a topic into a comparison table, baseline survey, benchmark roundup, or "what papers evaluated on which datasets with what scores and whether code is open".

Typical requests:

  • "Find recent baseline papers on agentic memory"
  • "What papers in the last month evaluated on dataset X?"
  • "Make me a markdown table of methods, datasets, and scores"

Goal

Turn a topic search into a structured markdown table:

  1. Search recent papers with deepxiv search
  2. Brief all candidates with deepxiv paper <id> --brief
  3. Keep the relevant papers, prioritizing papers with GitHub/code
  4. Inspect promising papers with deepxiv paper <id> --head
  5. Read experiment-related sections with deepxiv paper <id> --section ...
  6. Extract datasets, evaluation setup, and reported scores
  7. Write a markdown table summarizing the baselines

Default Workflow

1. Search by topic and date range

Use a broad search first.

deepxiv search "agentic memory" --date-from 2026-03-01 --limit 100 --format json

Default heuristics:

  • Use the user’s exact topic phrase first
  • Keep --limit high enough to avoid missing relevant papers
  • If results are noisy, refine the query with close variants

Examples:

deepxiv search "agentic memory" --date-from 2026-03-01 --limit 100 --format json
deepxiv search "memory agents long-horizon" --date-from 2026-03-01 --limit 100 --format json
deepxiv search "agent memory benchmark" --date-from 2026-03-01 --limit 100 --format json

2. Brief all candidates

For each arXiv ID, fetch:

deepxiv paper <arxiv_id> --brief

Capture:

  • title
  • arXiv ID
  • publish date
  • TLDR
  • keywords
  • GitHub URL
  • PDF/source URL

This is the screening step. Do not read full sections yet.

3. Filter and prioritize

Keep papers that are actually about the topic, not just adjacent terms.

Prioritize:

  • papers directly centered on the topic
  • empirical papers over purely conceptual ones
  • papers with GitHub/code
  • benchmark or comparison papers
  • papers with clear experiment sections

De-prioritize:

  • purely opinion or survey papers unless the user asked for surveys
  • papers with no clear evaluation evidence
  • papers only loosely related to the topic

If the list is still large, keep a primary set and a secondary set:

  • Primary: strongest and most relevant baselines
  • Secondary: adjacent or weaker evidence

4. Inspect paper structure

For retained papers:

deepxiv paper <arxiv_id> --head

Use --head to find experiment-bearing sections such as:

  • Experiments
  • Evaluation
  • Results
  • Benchmark
  • Main Results
  • Analysis

Also capture:

  • abstract
  • total token count
  • section names

5. Read only experiment-relevant sections

Once the right sections are known, read only those:

deepxiv paper <arxiv_id> --section Experiments
deepxiv paper <arxiv_id> --section Evaluation
deepxiv paper <arxiv_id> --section Results

Section selection guidance:

  • Start with Experiments or Evaluation
  • Read Results if the metrics are not clear
  • Read Introduction only if the task setup is still ambiguous
  • Read Appendix only if benchmark details are missing from the main paper

Avoid reading the entire paper unless the user explicitly asks for it.

Extraction Targets

For each retained paper, try to extract:

  • Title
  • arXiv ID
  • Paper URL
  • GitHub/code URL
  • Open-source status: Yes, No, or Unknown
  • Main task
  • Evaluation datasets / benchmarks
  • Key metrics
  • Best reported scores
  • Notes on experimental setting

If exact scores are not clearly available from the inspected sections:

  • leave the score field as Not clearly stated
  • do not invent or infer a number

If datasets are only partially visible:

  • include the datasets you can verify
  • mention that the list may be incomplete

Markdown Output

Write a markdown file in the workspace unless the user specifies another path.

Recommended filename:

<topic>-baseline-table-YYYY-MM-DD.md

Example:

agentic-memory-baseline-table-2026-04-01.md

Recommended output structure:

# Agentic Memory Baseline Table

Topic: agentic memory
Date range: 2026-03-01 to 2026-04-01
Search source: deepxiv

## Summary

- Number of search results
- Number of relevant papers retained
- Number with public code
- Main recurring datasets or benchmark families

## Baseline Table

| Title | arXiv | URL | Open Source | Code URL | Datasets / Benchmarks | Metrics / Scores | Notes |
| --- | --- | --- | --- | --- | --- | --- | --- |
| ... | ... | ... | ... | ... | ... | ... | ... |

## Inclusion Notes

- Which papers were excluded and why
- Which rows are based only on `--brief`
- Which rows were verified through `--head` and experiment/result sections

## Observations

- Common evaluation datasets
- Which papers appear strongest
- Where the benchmark story is still fragmented

Writing Rules

  • Prefer verified facts over broad summaries
  • Separate "paper is relevant" from "paper has strong benchmark evidence"
  • Be explicit when a row is missing score details
  • Mark open-source status conservatively
  • Keep the table compact but useful
  • Add short notes when comparisons are not apples-to-apples

Decision Rules

  • Always start with search and brief
  • Prefer papers with GitHub when deciding which ones to inspect first
  • Use --head before --section
  • Read only the sections needed to recover datasets and scores
  • If the topic is broad, tell the user when the table mixes multiple subtask types

Minimal Example

deepxiv search "agentic memory" --date-from 2026-03-01 --limit 100 --format json
deepxiv paper 2603.21489 --brief
deepxiv paper 2603.21489 --head
deepxiv paper 2603.21489 --section Experiments

Then write a markdown table with:

  • title
  • paper URL
  • open-source status
  • code URL
  • datasets / benchmarks
  • metrics / scores
  • short notes on what was actually verified

04 同仓库更多 Skills

pengqianhan
pengqianhan/AI-Human-Research-OS

workflow-skill-creator

Distills a completed user workflow or interaction into a reusable agent skill. Use when the user asks to turn their workflow, interaction, or multi-step process into a skill, or when they say "make this a skill", "create a skill from what we just did", "package this workflow" or similar. Do not use for creating skills from scratch without an existing workflow (use a generic skill-creator for that).

★ 3
pengqianhan
pengqianhan/AI-Human-Research-OS

uv

Checks whether the uv Python package manager is installed and installs it if missing. Ensures uv is on PATH. Use when another skill requires uv as a prerequisite.

★ 3
pengqianhan
pengqianhan/AI-Human-Research-OS

uniprot-database

Access protein metadata, function, taxonomy, and sequences across UniProtKB, UniParc, and UniRef. Use when searching for proteins, mapping identifiers, or retrieving functional annotations and publications. Don't use for sequence alignment, protein folding, or sequence similarity search (use specialized skills for those tasks).

★ 3
pengqianhan
pengqianhan/AI-Human-Research-OS

unibind-database

Queries the UniBind database for experimentally validated transcription factor (TF) binding sites. Use when retrieving direct TF-DNA interaction datasets, downloading binding site coordinates (BED/FASTA) for local analysis, or listing available datasets by species, cell line, or TF name. Don't use to query specific intervals, locations, genes, motif models or expression data.

★ 3
pengqianhan
pengqianhan/AI-Human-Research-OS

ucsc-conservation-and-tfbs

Fetch Evolutionary Conservation scores (phyloP, phastCons) and Transcription Factor Binding Sites (TFBS) from the UCSC Genome Browser. Use when analyzing whether genomic variants or regions are evolutionarily conserved, functionally important, or bounded by TF regulators across major projects (ENCODE, JASPAR, ReMap).

★ 3
pengqianhan
pengqianhan/AI-Human-Research-OS

string-database

Query the STRING database for protein-protein interactions (PPIs), functional enrichment, and homology. Use when the user asks about interactions between specific proteins, interaction evidence, confidence scores, protein interaction partners, or pathway enrichments.

★ 3