free-ai-token: An AI Money-Saving Research Skill with Real-Time Verification

Who It's For

• Individual users who want to save money on AI: anyone looking for currently available free AI API tokens, free AI products, or the cheapest AI memberships (including regional pricing), with neutral scoring and comparison.
• Works with any Agent Skills-compatible client (the official docs list Claude Code, Claude Desktop, QwenWork, Cursor, Codex, Gemini CLI, among others), provided the client has web access enabled (WebSearch/WebFetch) — the official documentation makes clear that real-time verification is the core of this skill, and it cannot work as designed without web access.
• Not a fit if: you want to run it in an offline environment; or you expect the system to automatically complete registration and payment — the official docs explicitly state that guided registration (references/auto-register.md) is optional and opt-in, and payment steps are never automated.
• The official docs do not state which languages are supported or guarantee how long specific free tiers last; these are unknowns, so rely on the real-time verification results within your session.

Get Your First Try Done in 3 Minutes

There is only one recommended main path: install via the Skills CLI into a compatible client.

1. Prerequisites: Node.js 18+ installed locally (npx depends on it), plus a client that is Agent Skills-compatible and has WebSearch/WebFetch enabled.
2. Install: run the official one-line command in your terminal:

npx skills add laowang-ai-xbb/free-ai-token

1. Trigger a low-risk task: after installation, type /deals in your client session. By the official definition, /deals is a one-shot scan of current free tiers and membership offers. In its default mode the skill performs read-only research: it verifies information one item at a time against vendors' official pages (when reachable), and every conclusion comes with a source and date. The official docs do not state whether this scan writes local files or produces other side effects — treat this as unknown; if your client requests any permission to write files or execute commands at runtime, review and approve before granting.
2. Success signal: you'll see the skill output numbered result cards (the presentation format described officially), each containing a neutral score, official link, source, and verification date.

Other official options: the official repository also lists /scan (which requires the host to provide a scheduler, for recurring watchlist scans) as a later, more advanced option; guided registration is a secondary, explicitly opt-in feature and isn't needed for your first try.

Use Cases

Case 1: One-shot scan of current free tiers and membership deals

Scenario: you want to know which AI products currently have free tiers you can take advantage of, and which membership is the best deal under regional pricing.

How: after installation, simply type /deals in your client — no parameters needed.

Result: the skill performs a market scan of current free tiers and membership offers, verifying in real time against each vendor's official pages, and outputs numbered result cards with official links, sources, and dates, ranked neutrally using a public scoring model. You can then decide for yourself which one to sign up for.

Case 2: Set up a recurring watchlist

Scenario: you don't just want to check once — you want to keep tracking changes to certain deals over time.

How: confirm your host client provides a scheduler, then use /scan.

Result: per the official description, /scan sets up a recurring watchlist scan that keeps monitoring deal changes. Note this is the only operation in this skill that requires scheduler support; if your host has no scheduler, the official docs offer no alternative — fall back to the one-shot approach in Case 1.

Usage Boundaries

• Web access: this is the core prerequisite. Every session visits vendors' official pages for real-time verification; if a page is unreachable, the official docs state verification happens only "when reachable," so conclusions may be incomplete.
• Side effects and confirmation points: read-only research by default. Guided registration only runs if you explicitly request it, every step is confirmed on screen, and payment steps are never automated — the payment action is always performed by you.
• Environment and client: npx requires Node.js 18+; the client must be Agent Skills-compatible with WebSearch/WebFetch enabled. The official docs list clients as "any …-compatible client (… and more)" without giving a closed compatibility list, and do not state which text or interface languages are supported.
• Officially stated limitations: the scoring model is neutral and public; all conclusions are based on real-time verification within the session, not memory.
• Unconfirmed information: the official evidence does not state whether the scan writes local files, whether it calls external services beyond search/fetch, or whether installation details are consistent across platforms. So don't assume it is "zero-risk" — if any permission is requested at runtime, confirm each item individually.