Trust & Safety
Last updated: 2026-09-24
Before you install anything, you should know exactly what happens. Here is an honest account of what our CLI does, what it explicitly does not do, and what we deliberately leave out.
What the CLI does
- Detects locally which AI tools are installed (or you pick them with --tool).
- Downloads only the skill files you unlocked, via a secured endpoint (token check).
- Places them tool-natively, e.g. Claude Code: ~/.claude/skills/…
- For Codex: updates only a marked block inside AGENTS.md. Your own content is kept.
- Open source (MIT): the full CLI code is public on GitHub. You can read exactly what the tool does.
- Verifies the delivered files against server-side SHA-256 checksums (integrity check).
- Asks before overwriting existing files in a terminal (with a notice on automated runs).
What the CLI does NOT do
- Reads no project files, no .env, no API keys.
- Sends no project content to us: only the token, skill name and detected tools leave your machine.
- Installs no background services and runs no foreign remote scripts.
- Writes only into the known skill paths: a double path guard (client and server) blocks writing outside them.
- Changes no global system settings.
Install without the CLI & remove again
You do not have to use the CLI. Every skill is downloadable as a ZIP and can be copied manually into your tool’s skill folder, the contents are shown as a file tree on each skill page before you get anything.
Remove again: locally, no token. Deletes the skill folder for each tool and, for Codex, removes only our marked AGENTS.md block:
npx --yes @skills-for-ai/cli remove <skill>Token & data
- One token per skill: valid only for that skill, usable an unlimited number of times.
- On a refund the token is revoked; after that it downloads nothing more.
- We store no project content. A download only increments the download counter.
- The endpoint runs over HTTPS; skill files live in private storage, not on a public CDN.
- A rate limit on the delivery endpoint curbs automated abuse.
Security badges: what they mean
Contains only instructions, templates and references, no executable scripts.
May contain executable scripts. Take a quick look before running. The file tree shows everything.
Runs entirely on your machine with your own AI: no external runtime, no running costs.
Does not call third-party services on its own.
SKILL.md following the open agentskills.io standard, portable and inspectable.
A fixed version + changelog per skill so updates stay traceable.
Compatibility per tool
Not every AI tool supports skills the same way. A skill is meant to be portable (SKILL.md), but each tool wires it in differently. Here is how to use it per tool:
| Tool | Skill support | Install path | Status |
|---|---|---|---|
| Claude Code | Native (auto-discovered, loads when needed) | CLI or folder ~/.claude/skills | Stable |
| Codex CLI | Native (open standard, call with $name) | CLI → ~/.agents/skills (agents also add a marked AGENTS.md block) | Stable |
| Cursor | Native (call with /name) | CLI → ~/.agents/skills (agents go into the project, where Cursor reads rules) | Stable |
| GitHub Copilot | Native (CLI, VS Code, JetBrains) | CLI → ~/.agents/skills (personal skills; not used by the cloud agent), agents → ~/.copilot/agents | Stable |
| Gemini CLI | Native (Agent Skills, activated when needed) | CLI → ~/.agents/skills (agents ship as an extension, then “gemini extensions install”) | Stable |
| Devin Desktop (formerly Windsurf) / Cline | Native (call with @name or /name) | CLI → ~/.codeium/windsurf/skills or ~/.cline/skills (Cline agents into the project, .clinerules/) | Stable |
| Claude (web & desktop app) | Skill upload (all plans, code execution on) | Upload the ZIP under “Customize” → “Skills” | Manual |
| ChatGPT (web) | Skill upload on Business, Enterprise and Edu plans | Upload the skill, otherwise paste the chat prompt | Manual |
As of 2026-09-24, checked against each vendor’s official docs. Tool support changes fast and we keep checking it. “Manual” means: it works, but without our CLI; you upload or paste the skill yourself.
The EU AI Act: who is responsible for what
The AI Act has applied since 2 August 2026; the transparency obligations in Article 50 have applied unchanged since then (the rules for high-risk systems were postponed to the end of 2027). Here is an honest account of what that means for this site, for our products, and for you.
No chatbot, no AI search, no AI feature: skills-for-ai runs no AI model and no inference. There is therefore no disclosure duty under Art. 50 (1) AI Act. If an AI chat is ever added, we will label it as such from the first interaction.
They are instruction and template files. They contain no model and compute nothing themselves. The AI system only comes about at your end, with your own AI.
Chatbot disclosure and machine-readable marking of generated content (Art. 50 (1) and (2)) apply to the provider of the AI service you use: e.g. Anthropic, OpenAI or Google.
If you publish AI-generated content, obligations of your own may apply: you must disclose deepfakes and certain texts of public interest (Art. 50 (4)), as well as emotion recognition and biometric categorisation (Art. 50 (3)).
Our products are not intended for high-risk applications under Annex III (incl. personnel selection, education, creditworthiness, law enforcement) and must not be used for the practices banned under Art. 5. See Terms § 6.
As of August 2026 · Supervisory authority in Germany: Bundesnetzagentur. Orientation, not legal advice: when in doubt, have your specific case checked.
Ready? The detailed step-by-step installation (with a manual alternative) is here: Go to the install guide