# arxiv-search

Adopt this skill: https://proskills.md/skills/other/ekkus93-vscode-skills-skills-arxiv-search

Page: https://proskills.md/skills/other/ekkus93-vscode-skills-skills-arxiv-search
This document is the public skill bundle for that page. It contains SKILL.md and every helper file the catalog carries. No login is required.

## File: skills/arxiv-search/SKILL.md
size: 4524
upstream: https://github.com/ekkus93/vscode_skills/blob/master/skills/arxiv-search/SKILL.md

````text
---
name: arxiv-search
description: Use when the user asks for arXiv papers on a topic, papers by a specific author, papers in an arXiv category, or a concise shortlist of relevant arXiv results with abstracts and links.
metadata: {"openclaw":{"os":["darwin","linux"],"requires":{"bins":["python3"]}}}
user-invocable: true
---

# arXiv Search

## Purpose

Use this skill to search arXiv and return a concise, source-linked summary of the most relevant papers.

This shared skill is self-contained. Use the helper script bundled with the skill instead of writing a new arXiv API request, XML parser, or result formatter by hand.

## Invocation

This skill is intended to be user-invocable by name.

If the runtime exposes skill slash commands, invoke it as:

- `/arxiv-search <query>`
- `/arxiv-search <query> --max-results 3`

Examples:

- `/arxiv-search transformer attention`
- `/arxiv-search author: Geoffrey Hinton`
- `/arxiv-search category: cs.LG`
- `/arxiv-search title: diffusion models`
- `/arxiv-search abstract: retrieval augmented generation`

If the query is missing, ask the user what they want to search for.

## When to use

- The user asks for recent arXiv papers on a topic.
- The user asks for papers by a specific author on arXiv.
- The user asks for arXiv results in a category such as `cs.LG`.
- The user wants a concise shortlist of relevant papers with abstracts and links.

## When not to use

- The request is not about arXiv or academic papers.
- The user needs full-text paper analysis rather than metadata-level search.
- The user needs general web research rather than arXiv specifically.

## Workflow

1. Determine the query from the user's request.
2. If the query is missing or ambiguous, ask a short clarifying question.
3. Run the bundled helper script with `python3`.
4. Return the helper output directly when it succeeds.
5. If the helper exits non-zero, treat stderr as the failure reason and report it plainly.
6. If no results are found, say so explicitly.
7. Do not guess paper titles, links, categories, or summaries.

## Commands

Basic topic search:

```bash
python3 "{baseDir}/arxiv_search.py" "transformer attention"
```

Author search:

```bash
python3 "{baseDir}/arxiv_search.py" "author: Geoffrey Hinton"
```

Category search:

```bash
python3 "{baseDir}/arxiv_search.py" "category: cs.LG"
```

Title or abstract search:

```bash
python3 "{baseDir}/arxiv_search.py" "title: diffusion models"
python3 "{baseDir}/arxiv_search.py" "abstract: retrieval augmented generation"
```

Limit the result count:

```bash
python3 "{baseDir}/arxiv_search.py" "transformer attention" --max-results 3
```

Check whether the required tool is installed:

```bash
command -v python3
```

Install `python3` on Ubuntu or Debian:

```bash
sudo apt-get update
sudo apt-get install -y python3
```

Install `python3` on macOS with Homebrew:

```bash
brew install python
```

## Input format

- Pass the query as one quoted argument.
- Supported query forms:
  - plain topic, for example `transformer attention`
  - `author: Geoffrey Hinton`
  - `category: cs.LG`
  - `title: diffusion models`
  - `abstract: retrieval augmented generation`
  - `all: retrieval augmented generation`
- Optional advanced form:
  - `python3 "{baseDir}/arxiv_search.py" "transformer attention" --max-results 3`
- Keep the query in one quoted string when using the shell command.

Valid example:

```bash
python3 "{baseDir}/arxiv_search.py" "author: Geoffrey Hinton"
```

Invalid example:

```bash
python3 "{baseDir}/arxiv_search.py" author: Geoffrey Hinton
```

## Output

Return the helper output directly when it succeeds.

Typical output includes:

- result count and sort mode
- numbered paper list
- title, authors, dates, category, short summary, abstract link, and PDF link
- confidence and freshness lines

## Constraints

- Use arXiv as the source for this skill.
- Use the bundled helper script instead of constructing a fresh API call and parser inline.
- `--max-results` must stay between 1 and 5.
- Do not dump raw XML.
- Do not guess paper metadata when arXiv returns no results or the request fails.
- If `python3` is missing, explain that the skill cannot run until it is installed.

## Missing Dependency Response

If `python3` is missing, prefer a short direct answer that includes install guidance, for example:

- `This skill needs python3. On Ubuntu or Debian, run: sudo apt-get update && sudo apt-get install -y python3`
- `This skill needs python3. On macOS with Homebrew, run: brew install python`
````

## File: README.md
upstream: https://github.com/ekkus93/vscode_skills/blob/master/README.md

````text
# Visual Studio Code Skills

![Visual Studio Code Skills](imgs/vscode.png)

This repository is a small workspace for experimenting with reusable agent skills, documenting how those skills are structured, and wiring them into a prompt-driven workflow.

The `vscode_skills` repository now contains both the repo-specific control files and the shared skills library.

The main repo content lives under `${HOME}/work/vscode_skills`, and the shared skill library now lives inside that repo at `${HOME}/work/vscode_skills/skills`.

For compatibility with older prompts and tooling, `${HOME}/skills` is a symlink that points to `${HOME}/work/vscode_skills/skills`.

## What This Project Does

This project gives you a place to:

- define reusable skills as folders with `SKILL.md` files
- keep a human-maintained skill index in `SKILL_LIST.md`
- route tasks through a prompt that tells the agent to choose and read a skill first
- experiment with OpenClaw-style skill authoring and invocation patterns
- add prompt-driven skill eval cases and grading helpers for Copilot-in-the-loop testing

In short, this repo now contains both the control layer and the content layer, with the shared skills stored in the in-repo `skills/` directory.

This repo can also hold skill-eval scaffolding that grades captured assistant outputs against case assertions when you want to test prompt-driven skill behavior.

## How It Works

There are four main pieces.

### 1. Repo instructions

This repo includes:

- `AGENTS.md`
- `.github/copilot-instructions.md`

These files tell the coding agent how to behave in this repository. They establish rules like:

- read relevant instructions before changing files
- consult the shared skills library when a task matches a reusable skill
- prefer small, explicit, low-risk changes

### 2. Prompt-driven skill selection

This repo includes a prompt file:

- `.github/prompts/use-a-skill.prompt.md`

Its job is simple: tell the agent to read the shared skill index, pick the closest matching skill, open that skill's `SKILL.md`, and follow it.

That means the prompt does not contain the skill logic itself. It delegates to the shared skill library.

### 3. Shared skill library

The real skills live in:

- `${HOME}/work/vscode_skills/skills`

That folder contains:

- `SKILL_LIST.md`: the index of officially available skills
- one folder per skill
- a `SKILL.md` inside each skill folder

For compatibility, `${HOME}/skills` points to the same directory via symlink.

Examples currently in the shared library include:

- `arxiv-search`
- `audio-transcribe`
- `bitcoin-price`
- `current-date-time`
- `docx-to-markdown`
- `pdf-to-markdown`
- `excel-to-delimited`
- `excel-to-markdown`
- `hacker-news-top10`
- `image-ocr`
- `net-diagnose-incident`
- `net-incident-intake`
- `net-client-health`
- `net-ap-rf-health`
- `net-dhcp-path`
- `net-dns-latency`
- `net-roaming-analysis`
- `net-auth-8021x-radius`
- `net-path-probe`
- `net-segmentation-policy`
- `net-ap-uplink-health`
- `net-stp-loop-anomaly`
- `net-l2-neighbor-discovery`
- `net-topology-map`
- `net-mac-path-trace`
- `net-subnet-inventory`
- `net-mdns-service-discovery`
- `net-gateway-health`
- `net-rf-interference-scan`
- `net-site-baseline-compare`
- `net-local-route-anomaly`
- `net-incident-correlation`
- `net-change-detection`
- `net-capture-trigger`
- `company-research`
- `news-search`
- `tweet-read`
- `stock-investment-review`
- `stock-research`
- `stock-review-market-context`
- `stock-review-output-contract`
- `stock-review-supporting-research`
- `weather`
- `wikipedia`
- `yahoo-finance`
- `yahoo-finance-cli`

#### NETTOOLS skills

The shared library also includes a dedicated NETTOOLS diagnostic set:

- `net-diagnose-incident`: top-level orchestrator that turns complaint intake into a structured diagnosis report and audit trail
- `net-incident-intake`: normalizes freeform complaints into NETTOOLS incident records
- `net-client-health`, `net-ap-rf-health`, `net-dhcp-path`, `net-dns-latency`, `net-ap-uplink-health`, `net-stp-loop-anomaly`: core diagnostic wrappers for client RF, AP RF, DHCP, DNS, AP uplinks, and L2 instability
- `net-roaming-analysis`, `net-auth-8021x-radius`, `net-path-probe`, `net-segmentation-policy`: deeper follow-up diagnostics for mobility, auth, path quality, and placement policy
- `net-l2-neighbor-discovery`, `net-topology-map`, `net-mac-path-trace`, `net-subnet-inventory`, `net-mdns-service-discovery`: topology and local-discovery helpers for adjacency, pathing, host inventory, and local services
- `net-gateway-health`, `net-rf-interference-scan`, `net-site-baseline-compare`, `net-local-route-anomaly`: follow-on topology and first-hop diagnostics for gateway health, RF overlap, baseline drift, and route anomalies
- `net-incident-correlation`, `net-change-detection`, `net-capture-trigger`: supporting investigation helpers for event correlation, change analysis, and gated capture planning

These wrappers rely on the shared runtime in `skills/nettools-core/`.

The important rule is that `SKILL_LIST.md` is the source of truth. A skill folder existing on disk is not enough by itself. If it is not listed in the index, it should not be treated as officially available.

### 4. Eval and test scaffolding

This repo can also hold prompt-driven skill eval scaffolding for VS Code and Copilot workflows.

That currently includes:

- `evals/cases/`: case files with prompts and assertion rules
- `evals/runner/skill_eval.py`: a small runner that grades captured assistant output
- `tests/test_skill_eval_runner.py`: unit tests for the runner itself

These evals are intentionally Copilot-in-the-loop. They grade captured output rather than launching GitHub Copilot automatically from a local CLI.

## Project Structure

Current repo structure:

```text
vscode_skills/
	.github/
		copilot-instructions.md
		prompts/
			use-a-skill.prompt.md
	AGENTS.md
	evals/
		cases/
			arxiv-search/
				arxiv-search-topic-success.json
		runner/
			skill_eval.py
	OPEN_CLAW_SKILL.md
	README.md
	skills/
		SKILL_LIST.md
		arxiv-search/
			audio-transcribe/
		company-research/
		pdf-to-markdown/
		excel-to-delimited/
		excel-to-markdown/
		net-ap-rf-health/
		net-ap-uplink-health/
		net-auth-8021x-radius/
		net-capture-trigger/
		net-change-detection/
		net-client-health/
		net-dhcp-path/
		net-diagnose-incident/
		net-dns-latency/
		net-incident-correlation/
		net-incident-intake/
		net-path-probe/
		net-roaming-analysis/
		net-segmentation-policy/
		net-stp-loop-anomaly/
		nettools-core/
		news-search/
		tweet-read/
		stock-investment-review/
		stock-research/
		yahoo-finance/
	tests/
		test_skill_eval_runner.py
```

Current shared skills structure:

```text
${HOME}/work/vscode_skills/skills/
	README.md
	SKILL_LIST.md
	arxiv-search/
		SKILL.md
		arxiv_search.py
		test_arxiv_search.py
	audio-transcribe/
		SKILL.md
		audio_transcribe.py
		test_audio_transcribe.py
	bitcoin-price/
		SKILL.md
	current-date-time/
		SKILL.md
	docx-to-markdown/
		SKILL.md
	pdf-to-markdown/
		SKILL.md
		pdf_to_markdown.py
		test_pdf_to_markdown.py
	excel-to-delimited/
		SKILL.md
		excel_to_delimited.py
		test_excel_to_delimited.py
	excel-to-markdown/
		SKILL.md
		excel_to_markdown.py
		test_excel_to_markdown.py
	hacker-news-top10/
		SKILL.md
	image-ocr/
		SKILL.md
	company-research/
		SKILL.md
		company_research.py
		test_company_research.py
	net-ap-rf-health/
		SKILL.md
	net-ap-uplink-health/
		SKILL.md
	net-auth-8021x-radius/
		SKILL.md
	net-capture-trigger/
		SKILL.md
	net-change-detection/
		SKILL.md
	net-client-health/
		SKILL.md
	net-dhcp-path/
		SKILL.md
	net-diagnose-incident/
		SKILL.md
	net-dns-latency/
		SKILL.md
	net-incident-correlation/
		SKILL.md
	net-incident-intake/
		SKILL.md
	net-path-probe/
		SKILL.md
	net-roaming-analysis/
		SKILL.md
	net-segmentation-policy/
		SKILL.md
	net-stp-loop-anomaly/
		SKILL.md
	nettools-core/
		README.md
	list-skills/
		SKILL.md
	news-search/
		SKILL.md
		news_search.py
		test_news_search.py
	tweet-read/
		SKILL.md
		t
````
