[What Are ICP Skills?]
/llms.txt, Caffeine agents get real, up-to-date ICP knowledge injected into every decision.> ICP now tops the AI & Big Data blockchain development ranking
As of July 13, 2026
The Internet Computer ranks #1 among blockchains for AI and big data application development — a clear signal that ICP is becoming the default chain for data-intensive, agentic workloads.
> 98.3 million transactions processed in a single day
July 8, 2026 — Chainspect data
On July 8, 2026, the Internet Computer processed a record 98.3 million transactions in 24 hours — the highest single-day throughput ever recorded on any blockchain per Chainspect. Proof that ICP doesn't just promise scale, it delivers it.
@ICPvibecoder loads ICP Skills guidelines at the very start of every session — before writing a single prompt. This habit primes the AI with ICP-specific context from prompt #1, cutting hallucinations and delivering better code quality instantly.
- >Eliminates made-up token standards and canister interfaces
- >Agents know the correct Motoko patterns, auth flows, and storage idioms
- >One line of cost, session-wide benefit — zero downside
▸ Paste this as your first message every session:
Load and follow the guidelines at: https://skills.internetcomputer.org/llms.txt> The JSON Option — More Precision, More Signal
Loading /llms.txt gives Caffeine a broad overview of every ICP capability — useful as a session anchor, but sometimes too wide. When you know exactly which capability you're building (HTTP outcalls, ICRC-1 tokens, stable memory…), paste in the individual skill JSON instead. Pure signal, zero noise — Caffeine responds faster and with higher precision to the exact patterns and APIs it needs.
- > 01Navigate to skills.internetcomputer.org and browse to the skill relevant to your current task.
- > 02Click the JSON option/button below that skill card.
- > 03Copy the full JSON content from the resulting page — it contains structured patterns, types, and best practices for that one capability.
- > 04Paste the JSON at the very start of your Caffeine prompt, followed immediately by your build instructions.
{
"name": "wallet-integration",
"title": "Wallet Integration",
"category": "Frontend",
"description": "Integrate ICP wallets (NFID, Plug, Stoic) into your app for authentication and transaction signing.",
"compatibility": {
"runtime": "Node.js",
"version": ">= 22"
},
"updated": "2025-01-15",
"urls": {
"html": "https://skills.internetcomputer.org/wallet-integration",
"markdown": "https://skills.internetcomputer.org/wallet-integration.md",
"json": "https://skills.internetcomputer.org/wallet-integration.json",
"source": "https://github.com/dfinity/icskills/blob/main/skills/wallet-integration"
},
"publisher": "DFINITY Foundation",
"canonicalRepo": "dfinity/icskills"
}Copy the full JSON block — not just the URL. Paste the entire object directly into your Caffeine prompt as context. The structured fields (description, patterns, compatibility) give the AI dense, machine-readable signal — a bare URL gives it almost nothing.
[Works for any skill on the list]
/llms.txt fetch for maximum focus.{
"skill": "http-outcalls",
"version": "1.0",
"patterns": [
"Use ic_cdk::api::management_canister::http_request",
"Always set max_response_bytes to bound costs",
"Use transform functions to canonicalise responses across replicas",
"Attach cycles: 230_949_972_000 for typical HTTPS GET"
],
"imports": ["use ic_cdk::api::management_canister::http_request::{...}"],
"notes": "HTTP outcalls require consensus — all replicas must return identical bytes. Use a transform function."
... [rest of skill JSON]
}
Now build me an app that calls the OpenAI API from a Motoko canister and
displays the response in a React frontend. Use the http_request management
canister API, attach sufficient cycles, and add a transform function so
responses are consensus-safe.> The Magic One-Liner
This single prompt supercharges every agent in the Caffeine pipeline with native ICP expertise:
Fetch https://skills.internetcomputer.org/llms.txt and follow its instructions when building on ICP.That's it. One line. It tells every agent to pull in the full ICP Skills reference before making any architectural or code decisions. The agents know to use the right Motoko patterns, the correct canister interfaces, and ICP-native authentication — automatically.
> Real Project Examples
🏦 DeFi Application
Without ICP Skills: agents make up token standards and ledger interfaces. With ICP Skills: ICRC-1/ICRC-2, proper ledger calls, token decimals — all correct first time.
Fetch https://skills.internetcomputer.org/llms.txt.
I'm building a DeFi app on ICP. Apply the relevant ICP Skills for token standards, ledger integration, and canister security.🔐 vetKeys Integration
vetKeys is a complex ICP primitive. The skill guides Caffeine to use the exact right API pattern for threshold key derivation — no hallucinated interfaces.
Fetch https://skills.internetcomputer.org/llms.txt.
I need vetKeys for encrypted data. Apply the vetKeys ICP Skill and follow the recommended pattern for key derivation and encryption.📜 Digital Will App
Time-locked data release using ICP-native features. ICP Skills ensures the canister timer setup, stable storage, and access control all follow best practices.
Fetch https://skills.internetcomputer.org/llms.txt.
Build a digital will app where assets are unlocked after a condition is met. Use canister timers, encrypted storage with vetKeys, and Internet Identity for authentication.> When to Use It
- Every new ICP project — make it a habit
- Before adding DeFi / token features
- Before implementing authentication flows
- When working with vetKeys, timers, or inter-canister calls
- When the backend needs to follow ICP security best practices
[Stack It With Your App Description]