Your skills, your account, your agents.
Skillfiles holds instructions your agents act on. Here is how access works, what the service keeps and where, and how to reach us.
OAuth, no API keys
Each agent signs in through OAuth, and you approve it in your browser after checking it is the agent you added. There are no long-lived keys to paste into config files.
Short-lived access
An agent’s access token lasts 15 minutes. The agent renews it with a refresh token, so you don’t sign in again, and every renewal checks that its approval still stands.
Revoke any agent
The connections page lists every agent you approved. Revoking one ends its approval at once, and any token it still holds stops working within 15 minutes at most.
One account, one library
Each account owns one library. Which library a request reaches comes only from a verified credential, and the library checks every caller against its own owner.
You review what agents write
Every skill an agent adds or changes waits for your review, with flags for added scripts, credential-like lines, new web hosts and downloads piped into a shell.
An optional copy on GitHub
On Pro, the library can back up every skill to a GitHub repository you own every hour. It is optional, and nothing else needs GitHub.
Only what the product needs.
Here is everything the service stores, and why. The service’s own logs get one short line per request, never an IP address, a query or skill content.
| Data | Why |
|---|---|
| Your skills: every file of every version, and the search cards and indexes built from them | To serve and search them, and to show you what changed |
| Your email address, a hash of your password and your browser sessions | To sign you in. A browser session lasts 30 days |
| The agents you approved and the tokens they hold | To know which agents act for you, and to let you revoke them |
| A record of each MCP request: the tool, the query and intent of a search, the agent’s software and version, its IP address and approximate location, and timing | For your dashboard. Kept for up to 400 days, never with file contents |
| Your review decisions | To know which version of each skill you approved |
| The GitHub repository you back up to, if you turn backup on | To push the hourly backup |
Where it lives
Skillfiles runs on Cloudflare. The accounts database, and every library created since accounts arrived with its files, are stored in Cloudflare’s EU jurisdiction.
Search cards, judging and review summaries use Claude Haiku, so skill text and search intents are sent to Anthropic’s API to produce them.
Full list of sub-processors: [sub-processor list]
Report a vulnerability
Found something? Email [security contact]. We acknowledge reports within [time] and credit researchers who want it.
Questions about access or data?
The privacy policy and the docs go into more detail.