Documentation
Installing skills
Every AgentHub download is a standard Agent Skills package. Pick your platform below. AgentHub never installs anything into your Claude account automatically — you stay in control.
Package format
A skill is a folder whose only required file is SKILL.md: YAML frontmatter with a name and description, followed by markdown instructions. Optional reference files, scripts and assets sit alongside it and are loaded only when needed.
my-skill/
├── SKILL.md # required
├── references/ # docs loaded on demand
│ └── guide.md
├── scripts/ # optional helpers
│ └── run.py
└── assets/ # templates, images---
name: my-skill
description: What it does. Use when …
license: Apache-2.0
metadata:
version: "1.0.0"
---
# My Skill
## Instructions
1. …
See [references/guide.md](references/guide.md).Claude apps (web & desktop)
Upload the skill ZIP in Claude's settings. Requires a plan with Skills and code execution enabled.
- 1Download
my-skill.zipfrom this page. Keep it zipped. - 2In Claude (web or desktop), open Settings → Capabilities and make sure code execution and file creation is turned on — skills depend on it.
- 3In the Skills section, choose Upload skill and select the ZIP.
- 4Make sure the skill is toggled on. Claude reads its description and loads it automatically when a request matches.
Skills in Claude apps require a paid plan; on Team and Enterprise plans an owner may need to enable Skills first. Menu labels can differ slightly between app versions. AgentHub cannot install skills into your Claude account for you.
Claude Code
Place the skill folder in ~/.claude/skills/ (personal) or .claude/skills/ (project).
- 1
Personal install (available in every project):
mkdir -p ~/.claude/skills unzip ~/Downloads/my-skill.zip -d ~/.claude/skills/ - 2
Or project install (shared with your team via git):
mkdir -p .claude/skills unzip ~/Downloads/my-skill.zip -d .claude/skills/ git add .claude/skills/my-skill - 3Confirm the layout is
skills/my-skill/SKILL.md, then start a new Claude Code session. Claude picks up the skill from its description; you can also invoke it directly with/my-skill.
Read bundled scripts before installing. Claude Code runs scripts with your local permissions when a skill calls them.
Claude API
Upload via the Skills endpoint and attach to requests with the code execution tool.
- 1
Unzip, then upload the skill folder with the Skills endpoint:
unzip my-skill.zip && cd my-skill args=() while IFS= read -r f; do args+=(-F "files[]=@$f;filename=my-skill/$f") done < <(find . -type f | sed 's|^\./||') curl https://api.anthropic.com/v1/skills \ -H "x-api-key: $ANTHROPIC_API_KEY" \ -H "anthropic-version: 2023-06-01" \ -H "anthropic-beta: skills-2025-10-02" \ -F "display_title=my-skill" "${args[@]}" - 2Attach the returned
skill_idin thecontainer.skillsparameter of a Messages request, alongside the code execution tool.
The Skills API is in beta; header names and parameters may change. Check the current Claude API documentation before deploying.
Claude Agent SDK
Load from the .claude/skills directory of your agent's working directory.
- 1
Unzip into your agent's project directory:
unzip my-skill.zip -d .claude/skills/ - 2Enable filesystem settings (project setting sources) and allow the Skill tool in your agent options so skills are discovered.
See the Claude Agent SDK documentation for the exact option names in your SDK language.
Other Agent Skills runtimes
Any agent that implements the open Agent Skills (SKILL.md) format.
- 1Unzip the package. It contains a single folder with
SKILL.mdat its root. - 2Copy the folder into your agent's skills directory as described by that runtime's documentation.
Agent Skills is an open format, but runtimes differ in which frontmatter fields and script types they support.
Safety
- Treat every skill as third-party code. Read SKILL.md and any scripts before installing.
- AgentHub validates structure and blocks executables, secrets and unsafe paths, and moderators review submissions — but review is not a guarantee.
- AgentHub never executes uploaded scripts. Claude apps and the API run skill scripts inside a sandboxed code-execution environment; Claude Code runs them on your machine with your permissions.
- Report anything suspicious from the skill page. Malware and prompt-injection reports are prioritised.