How much context do skills consume?

For 52,571 unique skills, the frontmatter median is 114 tokens, and the SKILL.md body median is 1,353 tokens, with the top 25% at 2,394 tokens or more. Installing 10 skills means about 1,100 tokens are always used before any input.

Are the install commands safe to use?

hotskills presents install commands pinned to a checked commit, and if a repository's security level changes, the date is displayed. The signals are automated and not a guarantee.

Large Language ModelsAI Safety & AlignmentOpen-Source AIZenn AI/ML7h agoNov 7, 2026, 19:01 JST

hotskills.dev: 77.6% of popular agent skills use network or shell

Send on LINE

3 Key Points

What happened
hotskills.dev tracks over 6,400 repositories and about 84,000 unique skills; of 849 popular repos checked, 77.6% use network calls or broad shell access, 12.3% have local scripts, and 10.1% are instructions only.
Why it matters
The site's stated purpose is to show what each skill executes before installation, since skills are often installed without inspection; 77.6% needing network or broad shell access suggests many are not self-contained, so reviewing content before install has clear value.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Get these free every morning

Context & Analysis

The project emerged from a simple gap: npm packages are often inspected before installation, but AI agent skills tend to be installed as-is. To address that, the developer collected candidates from GitHub code search, topic search, and awesome lists, judging whether a repository distributes skills from its file tree alone without using an LLM; git clone with --depth 1 --filter=blob:none --no-checkout fetches only the file list. Skills are counted by hashing the content of SKILL.md rather than by file count, and star counts are refreshed hourly via GitHub's GraphQL API.

The data also covers context cost. Across 52,571 unique skills, frontmatter that is always loaded into context has a median of 114 tokens, while the SKILL.md body loaded when a skill activates has a median of 1,353 tokens, with the top 25% at 2,394 tokens or more. That means 10 installed skills consume roughly 1,100 tokens before any input, and per-skill estimate pages let users choose lighter ones. Growth has been fast: in October, stars on skill-related repositories rose by 987,550 (+12.0%) and 513 new repositories appeared.

To address the fact that skills can change after installation, the install commands hotskills presents are pinned to checked commits. A read-only MCP server is also published on the official MCP Registry (dev.hotskills/hotskills), returning rankings, security signals, and commit-pinned install commands for free.

FAQ

How much context do skills consume?: For 52,571 unique skills, the frontmatter median is 114 tokens, and the SKILL.md body median is 1,353 tokens, with the top 25% at 2,394 tokens or more. Installing 10 skills means about 1,100 tokens are always used before any input.
How fast are skill repositories growing?: In October alone, stars on skill-related repositories rose by 987,550 (+12.0%), and 513 new repositories appeared.
Are the install commands safe to use?: hotskills presents install commands pinned to a checked commit, and if a repository's security level changes, the date is displayed. The signals are automated and not a guarantee.

Zenn AI/MLRead Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.