Documentation
Everything you need to know about SkillUp and Observatory.
Marketplace
What is SkillUp?
SkillUp is a marketplace for reusable Claude Code skills. Skills are structured workflows that teach Claude how to perform specific tasks — from running test suites to debugging Docker containers to auditing codebases for security issues.
Makers publish skills to the marketplace, and anyone can browse, download, or fork them. Skills are installed as slash commands in Claude Code, so you can trigger them with a simple /skill-name in your terminal.
Getting started
- Create an account — Sign up at the top right. Your username is unique and used in skill URLs.
- Explore skills — Browse trending skills, search by name or category, and discover what other makers have published.
- Download or fork — Download a skill as a zip, or use Observatory to install it directly into Claude Code.
- Publish your own — Write a SKILL.md, connect Observatory, and push to the marketplace.
Exploring & downloading skills
The Explore page shows trending skills, most followed makers, and recently added skills. You can filter by category (coding, devops, data, design, writing, content, research, tools) and sort by downloads or recency.
Every skill has a detail page with three tabs:
- Preview — Rendered markdown showing what the skill does and how to use it.
- Files — Syntax-highlighted source code for every file in the skill.
- Comments — Community discussion and feedback.
Downloading: Click the Download button to get a zip file containing the SKILL.md and all supporting files.
Forking via Observatory: Click the Fork button to reveal the Observatory install command. Copy it, paste it into Claude Code, and the skill is installed locally.
Publishing a skill
To publish a skill to the marketplace you need Observatory connected to your account. Then:
- Write a
SKILL.mdfile with YAML frontmatter (name, description, category tags). - Add any supporting scripts or templates alongside it.
- Run
/observatory push path/to/SKILL.mdin Claude Code.
Observatory validates your metadata (name must exist, description must be 20+ characters), collects all files in the skill directory, and publishes them to the marketplace. Your skill appears on your dashboard and is immediately searchable.
Following makers
Follow other makers to keep track of their work. The Follow button appears on every user profile. Your Following page shows everyone you follow with their follower and skill counts.
You can also search for users by username or display name from the Following page. The Explore page features a "Most Followed Makers" section to help you discover active publishers.
API tokens & authentication
Your Account page shows two tokens:
- Access token — A JWT used to authenticate requests. Expires after 1 hour.
- Refresh token — Used to get a new access token when the current one expires. Longer-lived.
Observatory stores both tokens locally and handles refresh automatically. You don't need to manage token rotation manually — Observatory takes care of it.
Observatory Skill
What is Observatory?
Observatory is a Claude Code skill that connects your local development environment to the SkillUp marketplace. It lives inside Claude Code as a slash command, silently tracking your workflows and helping you turn recurring patterns into reusable skills.
It operates in two modes:
- Passive (stop hook) — Runs silently after every conversation turn. Records tool calls, groups them into workflow blocks, and saves session history.
- Active (slash commands) — Run
/observatoryto interact with the marketplace, detect patterns, and manage skills.
Installing Observatory
- Run the install script in your terminal:
curl -fsSL https://skillup-market.vercel.app/install.sh | bash - Go to your Account page and copy your access token and refresh token.
- In Claude Code, run:
/observatory login <access-token> <refresh-token> - You're connected. Observatory will start tracking your sessions and you can fetch, push, and fork skills from the marketplace.
Commands
| Command | Description |
|---|---|
/observatory help | Show help and current session stats |
/observatory detect | Analyze session history for recurring workflow patterns and suggest skills to create |
/observatory fetch <user/skill> | Download and install a skill from the marketplace with safety review |
/observatory push <path> | Publish a skill to the marketplace |
/observatory fork <user/skill> | Fork a skill to a local directory for editing |
/observatory login <token> [refresh] | Store your marketplace credentials locally |
File structure
~/.claude/skills/observatory/
├── SKILL.md # Skill definition & slash command registration
├── scripts/
│ ├── engine.py # Core brain & command router
│ ├── parser.py # Transcript reader
│ ├── auth.py # Auth helper with token refresh
│ ├── fetcher.py # Marketplace fetch + safety review
│ ├── publisher.py # Marketplace publish
│ ├── fork.py # Marketplace fork
│ └── login.py # Token storage
└── templates/
├── skill_template.md # SKILL.md template with placeholders
└── examples.md # Example generated skillsengine.py is the central hub (~685 lines). As a stop hook it processes transcripts, groups tool calls into workflow blocks, filters noise, collapses repetitive calls, categorizes files and commands, computes fingerprints, and saves to session history. As a command router it delegates to the other scripts.
parser.py reads Claude Code's .jsonl transcript files and extracts structured events — user intents and tool calls with file paths and input summaries.
auth.py provides shared authentication. It stores credentials in ~/.observatory/config.json, handles token refresh automatically, and provides an authed_request() function that retries on 401.
fetcher.py implements a two-step install process. First it runs a safety pre-scan checking for prompt injection, dangerous commands, and obfuscation patterns. If it passes, the full content is shown for review. Only after explicit confirmation (with --force) is the skill written to disk.
Local data storage
| Path | Purpose |
|---|---|
~/.observatory/config.json | Marketplace credentials (token, refresh_token, api_url) |
~/.observatory/pending.json | Latest session's workflow blocks awaiting analysis |
~/.observatory/history.json | Last 50 sessions of workflow blocks for cross-session pattern matching |
~/.observatory/forks/ | Locally forked skills for editing before re-publishing |
Safety & security
Observatory includes a built-in safety layer when fetching skills from the marketplace:
- Pre-scan — Heuristic checks for prompt injection phrases, dangerous shell commands (rm -rf, curl | bash, chmod 777), and obfuscation patterns (base64 decode + eval, excessive hex escapes).
- Content review — If the pre-scan passes, the full skill content is displayed for line-by-line review before installation.
- Explicit confirmation — Skills are only written to disk after you confirm with the
--forceflag.
Usernames are unique across the platform, so skill references like username/skill-name are unambiguous.