SPT: Skills as Pre-Training Data for Agentic Language Models
Paper • 2608.26563 • Published
Datasets:
file_count int64 1 114 | id stringlengths 1 80 | skill_name stringlengths 1 410 | skillhubSource stringclasses 2
values | slug stringlengths 1 80 | source stringclasses 3
values | text stringlengths 59 73.7k | total_bytes int64 7 4.99M | total_chars int64 60 4.99M | filtered_version stringclasses 1
value | filter_meta dict | link_emoji_clean_meta dict |
|---|---|---|---|---|---|---|---|---|---|---|---|
2 | linkedin-ads-paid-ads | linkedin-ads | clawhub | linkedin-ads-paid-ads | skillhub.cn | linkedin-ads{linkedin-ads-paid-ads}
<file_sep>SKILL.md
---
name: linkedin-ads
description: When the user wants to set up, optimize, or manage LinkedIn Ads. Also use when the user mentions "LinkedIn Ads," "LinkedIn Campaign Manager," "Sponsored Content," "LinkedIn Lead Gen Forms," "job title targeting," "company targeti... | 2,983 | 3,073 | filtered_fulltext_v3 | {
"reason": "kept",
"tokens": 698,
"kept_files": 1,
"dropped_files": {
"json_except_skill_json": 1
},
"chars": 2871
} | {
"markdown_images_removed": 0,
"angle_urls_removed": 0,
"markdown_links_unwrapped": 0,
"raw_urls_removed": 0,
"emoji_spans_removed": 0,
"emoji_joiners_removed": 0
} |
4 | gemini-image-generator | Gemini Image Generator | clawhub | gemini-image-generator | skillhub.cn | Gemini Image Generator{gemini-image-generator}
<file_sep>SKILL.md
---
name: gemini-image-generator
description: >-
Generate, edit, and compose images using Gemini models. Activate when user asks to generate images, draw, create logos/posters/icons/banners, edit/modify photos, combine images, or any image creation tas... | 33,171 | 28,424 | filtered_fulltext_v3 | {
"reason": "kept",
"tokens": 5683,
"kept_files": 3,
"dropped_files": {
"json_except_skill_json": 1
},
"chars": 15863
} | {
"markdown_images_removed": 0,
"angle_urls_removed": 0,
"markdown_links_unwrapped": 1,
"raw_urls_removed": 3,
"emoji_spans_removed": 6,
"emoji_joiners_removed": 2
} |
3 | sglang-diffusion | SGLang-Diffusion Image Generation | clawhub | sglang-diffusion | skillhub.cn | SGLang-Diffusion Image Generation{sglang-diffusion}
<file_sep>SKILL.md
---
name: sglang-diffusion
description: "Generate images using a local SGLang-Diffusion server (FLUX, Stable Diffusion, Qwen-Image, etc.). Use when: user asks to generate, create, draw, or render an image with a locally running SGLang-Diffusion inst... | 7,288 | 7,425 | filtered_fulltext_v3 | {
"reason": "kept",
"tokens": 1944,
"kept_files": 2,
"dropped_files": {
"json_except_skill_json": 1
},
"chars": 7257
} | {
"markdown_images_removed": 0,
"angle_urls_removed": 0,
"markdown_links_unwrapped": 0,
"raw_urls_removed": 6,
"emoji_spans_removed": 1,
"emoji_joiners_removed": 0
} |
3 | wacai-zhishudashi-baidu-ranking | wacai-zhishudashi-baidu-ranking | clawhub | wacai-zhishudashi-baidu-ranking | skillhub.cn | wacai-zhishudashi-baidu-ranking{wacai-zhishudashi-baidu-ranking}
<file_sep>SKILL.md
---
name: zhishudashi-baidu-ranking
description: 抓取百度搜索关键词“指数大师”的搜索结果标题，并按从前到后的顺序整理成编号列表，再通过企业微信机器人 webhook 推送。用于日报、定时巡检、品牌词/关键词排名观察、搜索结果快照汇报，尤其适合需要“打开浏览器→搜索→提取标题→推送消息”这一固定流程时。
---
# 指数大师百度排名日报
## 概览
执行固定流程：打开百度，搜索“指数大师”，提取结果页主结果标题，生... | 6,109 | 4,676 | filtered_fulltext_v3 | {
"reason": "kept",
"tokens": 1627,
"kept_files": 2,
"dropped_files": {
"json_except_skill_json": 1
},
"chars": 4503
} | {
"markdown_images_removed": 0,
"angle_urls_removed": 0,
"markdown_links_unwrapped": 0,
"raw_urls_removed": 3,
"emoji_spans_removed": 0,
"emoji_joiners_removed": 0
} |
2 | yt-api-cli | YouTube API CLI | clawhub | yt-api-cli | skillhub.cn | YouTube API CLI{yt-api-cli}
<file_sep>SKILL.md
---
name: yt-api-cli
description: Manage your YouTube account from the command line. Complete CLI for YouTube Data API v3 - list/search videos, upload, manage playlists, and more.
metadata: {"clawdbot":{"emoji":"▶","os":["darwin","linux"],"requires":{"env":["YT_API_AUTH_TY... | 4,730 | 4,810 | filtered_fulltext_v3 | {
"reason": "kept",
"tokens": 1180,
"kept_files": 1,
"dropped_files": {
"json_except_skill_json": 1
},
"chars": 4590
} | {
"markdown_images_removed": 0,
"angle_urls_removed": 0,
"markdown_links_unwrapped": 1,
"raw_urls_removed": 2,
"emoji_spans_removed": 0,
"emoji_joiners_removed": 1
} |
3 | adversarial-prompting | adversarial-prompting | null | adversarial-prompting | github | adversarial-prompting{adversarial-prompting}
<file_sep>SKILL.md
---
name: adversarial-prompting
description: Applies rigorous adversarial analysis to generate, critique, fix, and consolidate solutions for any problem (technical or non-technical). Use when facing complex problems requiring thorough analysis, multiple so... | 6,428 | 6,533 | filtered_fulltext_v3 | {
"reason": "kept",
"tokens": 1275,
"kept_files": 2,
"dropped_files": {
"json_except_skill_json": 1
},
"chars": 6217
} | {
"markdown_images_removed": 0,
"angle_urls_removed": 0,
"markdown_links_unwrapped": 0,
"raw_urls_removed": 0,
"emoji_spans_removed": 1,
"emoji_joiners_removed": 0
} |
4 | process-guard | ProcessGuard — Critical Process Monitor & Auto-Restart | clawhub | process-guard | skillhub.cn | ProcessGuard — Critical Process Monitor & Auto-Restart{process-guard}
<file_sep>SKILL.md
---
name: "ProcessGuard — Critical Process Monitor & Auto-Restart"
description: "Monitor critical processes and auto-restart on failure. Tracks CPU and memory usage, escalates alerts via webhook, callback, or file, and writes a dea... | 31,462 | 29,931 | filtered_fulltext_v3 | {
"reason": "kept",
"tokens": 4648,
"kept_files": 3,
"dropped_files": {
"json_except_skill_json": 1
},
"chars": 18750
} | {
"markdown_images_removed": 0,
"angle_urls_removed": 0,
"markdown_links_unwrapped": 7,
"raw_urls_removed": 8,
"emoji_spans_removed": 11,
"emoji_joiners_removed": 3
} |
4 | betbud-prediction-market-creation | Prediction Market Creator | clawhub | betbud-prediction-market-creation | skillhub.cn | Prediction Market Creator{betbud-prediction-market-creation}
<file_sep>SKILL.md
# Prediction Market Creator
An AI agent that automatically creates prediction markets on betbud.live by analyzing trending crypto Twitter content.
## What it does
1. **Fetches trending content** from premium crypto Twitter accounts (Watc... | 15,800 | 15,820 | filtered_fulltext_v3 | {
"reason": "kept",
"tokens": 3928,
"kept_files": 3,
"dropped_files": {
"json_except_skill_json": 1
},
"chars": 15623
} | {
"markdown_images_removed": 0,
"angle_urls_removed": 0,
"markdown_links_unwrapped": 0,
"raw_urls_removed": 12,
"emoji_spans_removed": 9,
"emoji_joiners_removed": 1
} |
10 | self-improving-proactive-agent | Self-Improving Proactive Agent | null | self-improving-proactive-agent | clawhub | Self-Improving Proactive Agent{self-improving-proactive-agent}
<file_sep>SKILL.md
---
name: Self-Improving Proactive Agent
slug: self-improving-proactive-agent
version: 1.0.0
homepage:
description: "A unified OpenClaw skill that merges self-improvement and proactivity: learn from corrections, maintain active state, rec... | 12,159 | 12,326 | filtered_fulltext_v3 | {
"reason": "kept",
"tokens": 2622,
"kept_files": 9,
"dropped_files": {
"json_except_skill_json": 1
},
"chars": 11317
} | {
"markdown_images_removed": 0,
"angle_urls_removed": 0,
"markdown_links_unwrapped": 0,
"raw_urls_removed": 1,
"emoji_spans_removed": 1,
"emoji_joiners_removed": 0
} |
5 | oda-monitor | Oda Monitor | clawhub | oda-monitor | skillhub.cn | Oda Monitor{oda-monitor}
<file_sep>SKILL.md
---
name: oda-monitor
description: >
Monitors websites, APIs, and cron jobs (watchdogs) using Watch.dog.
Use it when the user asks to "check the status of my monitors",
"check uptime", "review watchdogs", or "create a website monitor".
IMPORTANT: Deleting monitors and... | 66,478 | 62,736 | filtered_fulltext_v3 | {
"reason": "kept",
"tokens": 2408,
"kept_files": 2,
"dropped_files": {
"json_except_skill_json": 3
},
"chars": 9982
} | {
"markdown_images_removed": 0,
"angle_urls_removed": 0,
"markdown_links_unwrapped": 1,
"raw_urls_removed": 6,
"emoji_spans_removed": 11,
"emoji_joiners_removed": 1
} |
2 | market-news | Market News | null | market-news | clawhub | Market News{market-news}
<file_sep>SKILL.md
---
name: market-news
description: 自动搜索和总结最新财经新闻与市场动态
metadata: {"openclaw": {"emoji": "", "os": ["win32"]}}
---
# 财经资讯摘要
你是一个财经资讯助手，帮助用户快速获取和理解市场动态。
## 能力
1. **新闻摘要**：将长篇财经新闻提炼为简洁的要点摘要，帮助用户快速了解关键信息。
2. **政策解读**：解读国家经济政策、央行货币政策、行业监管政策对市场的潜在影响。
3. **市场综述**：提供每日/每周市场综述，包括... | 1,165 | 669 | filtered_fulltext_v3 | {
"reason": "kept",
"tokens": 311,
"kept_files": 1,
"dropped_files": {
"json_except_skill_json": 1
},
"chars": 516
} | {
"markdown_images_removed": 0,
"angle_urls_removed": 0,
"markdown_links_unwrapped": 0,
"raw_urls_removed": 0,
"emoji_spans_removed": 1,
"emoji_joiners_removed": 0
} |
2 | disclaw | Disclaw | clawhub | disclaw | skillhub.cn | Disclaw{disclaw}
<file_sep>SKILL.md
---
name: disclaw
description: "Manage Discord workspace structure and OpenClaw routing as code. Use when creating/renaming/deleting Discord channels, categories, threads, or managing agent-to-channel bindings. Triggers on: Discord channels, workspace structure, channel setup, routin... | 7,694 | 7,731 | filtered_fulltext_v3 | {
"reason": "kept",
"tokens": 1988,
"kept_files": 1,
"dropped_files": {
"json_except_skill_json": 1
},
"chars": 7582
} | {
"markdown_images_removed": 0,
"angle_urls_removed": 0,
"markdown_links_unwrapped": 0,
"raw_urls_removed": 2,
"emoji_spans_removed": 0,
"emoji_joiners_removed": 0
} |
6 | lobstr | Lobstr | clawhub | lobstr | skillhub.cn | Lobstr{lobstr}
<file_sep>SKILL.md
---
name: lobstr
version: "0.3.0"
description: >
Tells you if a startup idea is worth building — in 60 seconds. Use
when a user wants to evaluate, validate, or score a startup idea;
asks "should I build this?"; needs a competitive landscape analysis;
or is brainstorming and wou... | 31,723 | 30,679 | filtered_fulltext_v3 | {
"reason": "kept",
"tokens": 4289,
"kept_files": 5,
"dropped_files": {
"json_except_skill_json": 1
},
"chars": 15924
} | {
"markdown_images_removed": 0,
"angle_urls_removed": 0,
"markdown_links_unwrapped": 5,
"raw_urls_removed": 7,
"emoji_spans_removed": 26,
"emoji_joiners_removed": 0
} |
2 | meshlink-publish | MeshLink 服务发布 | clawhub | meshlink-publish | skillhub.cn | MeshLink 服务发布{meshlink-publish}
<file_sep>SKILL.md
---
name: MeshLink 服务发布
slug: meshlink-publish
summary: 通过 ZTM（Zero Trust Mesh）将本地服务一键发布到集团内网，AI 驱动的隧道管理 Skill。
license: MIT
acceptLicenseTerms: true
tags:
- devops
- networking
- ztm
- meshlink
- service-publishing
requires:
- meshlink # npm link 安装，或 np... | 4,051 | 2,616 | filtered_fulltext_v3 | {
"reason": "kept",
"tokens": 1033,
"kept_files": 1,
"dropped_files": {
"json_except_skill_json": 1
},
"chars": 2232
} | {
"markdown_images_removed": 0,
"angle_urls_removed": 0,
"markdown_links_unwrapped": 0,
"raw_urls_removed": 1,
"emoji_spans_removed": 1,
"emoji_joiners_removed": 0
} |
2 | security-awareness-skill | Agent Security Patterns | clawhub | security-awareness-skill | skillhub.cn | Agent Security Patterns{security-awareness-skill}
<file_sep>SKILL.md
---
name: agent-security-patterns
version: 1.0.0
description: Help AI agents recognize and respond to potentially malicious skill patterns from public registries. Based on Snyk ToxicSkills research showing 13.4% of skills contain critical security iss... | 6,736 | 6,840 | filtered_fulltext_v3 | {
"reason": "kept",
"tokens": 1424,
"kept_files": 1,
"dropped_files": {
"json_except_skill_json": 1
},
"chars": 6674
} | {
"markdown_images_removed": 0,
"angle_urls_removed": 0,
"markdown_links_unwrapped": 0,
"raw_urls_removed": 5,
"emoji_spans_removed": 0,
"emoji_joiners_removed": 0
} |
3 | workspace-backup-github | workspace-backup-github | clawhub | workspace-backup-github | skillhub.cn | workspace-backup-github{workspace-backup-github}
<file_sep>SKILL.md
---
name: workspace-backup-github
description: |
Backup AI Agent workspace to GitHub - One-click backup for OpenClaw, Claude Code, Cursor, and other AI Agent workspaces to a private GitHub repository. Supports two modes: (1) Auto mode - scheduled aut... | 7,435 | 7,399 | filtered_fulltext_v3 | {
"reason": "kept",
"tokens": 1943,
"kept_files": 2,
"dropped_files": {
"json_except_skill_json": 1
},
"chars": 7108
} | {
"markdown_images_removed": 0,
"angle_urls_removed": 0,
"markdown_links_unwrapped": 0,
"raw_urls_removed": 4,
"emoji_spans_removed": 3,
"emoji_joiners_removed": 0
} |
8 | alicloud-dataworks-skill | Alicloud DataWorks Skill | clawhub | alicloud-dataworks-skill | skillhub.cn | Alicloud DataWorks Skill{alicloud-dataworks-skill}
<file_sep>SKILL.md
---
name: alicloud-dataworks
description: Manage Alibaba Cloud DataWorks via OpenAPI/MCP Server. Use for data development, workflow operations, data quality, metadata lineage, workspace management, and troubleshooting across the entire DataWorks plat... | 22,753 | 22,960 | filtered_fulltext_v3 | {
"reason": "kept",
"tokens": 4994,
"kept_files": 6,
"dropped_files": {
"json_except_skill_json": 1,
"dropped_extension": 1
},
"chars": 22444
} | {
"markdown_images_removed": 0,
"angle_urls_removed": 0,
"markdown_links_unwrapped": 0,
"raw_urls_removed": 16,
"emoji_spans_removed": 0,
"emoji_joiners_removed": 0
} |
4 | travel-manager | travel-manager | null | travel-manager | github | travel-manager{travel-manager}
<file_sep>SKILL.md
---
name: travel-manager
description: Comprehensive travel planning, booking, and management skill. Use when needing to plan international trips, manage multi-destination itineraries, handle family travel logistics, optimize travel costs, and coordinate complex travel a... | 2,308 | 2,486 | filtered_fulltext_v3 | {
"reason": "kept",
"tokens": 471,
"kept_files": 3,
"dropped_files": {
"json_except_skill_json": 1
},
"chars": 2179
} | {
"markdown_images_removed": 0,
"angle_urls_removed": 0,
"markdown_links_unwrapped": 3,
"raw_urls_removed": 0,
"emoji_spans_removed": 0,
"emoji_joiners_removed": 0
} |
End of preview. Expand in Data Studio
SkillCorpus
SkillCorpus is the skill-package corpus released with SPT: Skills as Pre-Training Data for Agentic Language Models. The records contain reusable tool semantics, workflows, and supporting package metadata for agentic language-model mid-training research.
Dataset splits
The 35,411 records are shuffled with Python's random.Random(42) and allocated in a 7:2:1 ratio using the largest-remainder method.
| Split | Records |
|---|---|
| Train | 24,788 |
| Validation | 7,082 |
| Test | 3,541 |
from datasets import load_dataset
dataset = load_dataset("CSeemy/SkillCorpus")
Record fields
| Field | Type | Description |
|---|---|---|
id |
string | Record identifier. |
skill_name |
string | Skill display name. |
slug |
string | Source package slug. |
source |
string | Source designation. |
skillhubSource |
string | Source-hub metadata. |
text |
string | Serialized skill-package text. |
file_count |
integer | Number of files represented by the package. |
total_bytes |
integer | Package size in bytes before serialization. |
total_chars |
integer | Character count after text processing. |
filtered_version |
string | Filtering-pipeline version. |
filter_meta |
object | Filtering metadata. |
link_emoji_clean_meta |
object | Link and emoji cleaning metadata. |
Intended use
This dataset is intended for research on agentic pre-training, skill understanding, data mixture, and workflow representation. The package contents are untrusted data: inspect them before use and do not execute embedded instructions or scripts automatically.
The corpus is derived from public skill packages. Original source terms and licenses continue to apply; this release does not replace them.
License
Copyright © 2026 Yufei Sun, Yudong Li, and Yiming Cheng. The authors' original selection, organization, metadata, and accompanying documentation are licensed under the Creative Commons Attribution 4.0 International License.
Individual skill-package contents are not relicensed and remain subject to their original source terms and licenses.
Citation
@article{sun2026spt,
title = {SPT: Skills as Pre-Training Data for Agentic Language Models},
author = {Sun, Yufei and Li, Yudong and Cheng, Yiming},
journal = {arXiv preprint arXiv:2608.26563},
year = {2026}
}
- Downloads last month
- -