Agent skill
The Chord skill teaches an agent to use Chord the way a careful trader’s assistant would: read prices, boards, candles, holders and wallet PnL on all eight EVM chains, quote a swap across every venue, and prepare an unsigned transaction, without ever signing anything itself. It follows the open Agent Skills format, so Claude Code, Codex, OpenClaw, Hermes and other agents load it the same way.
Install
Section titled “Install”One command, no repository access needed. It unpacks the whole skill (SKILL.md, references, the client) into the folder your agent reads:
mkdir -p ~/.claude/skills/chord && curl -fsSL https://getchord.xyz/.well-known/agent-skills/chord.tar.gz | tar -xz -C ~/.claude/skills/chordOr with the skills CLI, which reads the site’s discovery index (-a picks the agent; -g installs for your user instead of the project):
npx skills add https://getchord.xyz -s chord -a claude-codeTools that follow Agent Skills discovery find it from the domain alone: https://getchord.xyz/.well-known/agent-skills/index.json lists the archive with its SHA-256 digest, so a client can check the bytes before loading them. SKILL.md by itself is at https://getchord.xyz/skill.md, and every file is also under /docs/skills/chord/ (table below).
Then, from that folder, bash scripts/install.sh copies it to ~/.agents/skills/chord as well and prints the paths for OpenClaw (~/.openclaw/skills/chord) and Hermes (~/.hermes/skills/chord). For one project only: bash scripts/install.sh --project /path/to/project (writes .claude/skills/chord there). The installer is idempotent and only replaces a folder that already holds this skill.
Optional: export CHORD_API_KEY=... raises the rate limit and unlocks Chord Ask. Data calls work without a key at 60 requests a minute. Create a key at getchord.xyz/developer, or in Ask at getchord.xyz/ask (sign in, Settings, API Key). Both are the same kind of key and work for data and Ask.
What it covers
Section titled “What it covers”| Ask your agent | What the skill does |
|---|---|
| “Price of NVDA’s stock token and its gap to the real stock” | reads the Robinhood Chain stock screener: last on-chain price, reference price and its source, gap, off-hours share |
| “Quote 50 USDC to AERO on Base” | resolves both tokens on the chain, converts the amount exactly, quotes every venue, shows fee, minimum received, impact and route |
| “Which tokens on Arc traded the most in the last 24h?” | reads the chain board, flags wash and impostor rows and look-alike tickers |
| “What’s moving on pump.fun? Who bought this Solana token?” | reads Chord’s Solana tape (read-only): board, token stats, traders, wallet PnL, with base58 mints kept exactly as given and the tape’s gaps (no prop AMMs or order books) stated |
| “Who is buying ARGUS, where is support?” | asks Chord Ask (needs a key) and relays the answer with its sources |
| “Swap it” | shows a full confirmation summary, waits for an explicit yes in a new message, then builds the unsigned transaction for the user’s own wallet |
Chains: Ethereum (1), Arbitrum One (42161), Base (8453), BNB Chain (56), Avalanche (43114), Monad (143), Robinhood Chain (4663), Arc (5042), plus Solana: data, and swaps through /v1/sol/trade/* (the user’s wallet signs).
The safety model
Section titled “The safety model”- Read and prepare, never sign. The skill never signs, broadcasts or asks for a private key. The swap transaction is unsigned and goes to the user’s own wallet.
- Two-step confirmation. The quote and summary come in one message; the transaction is built only after the user says yes in the next one. Any change to amount, token, chain, slippage or wallet voids the confirmation. The bundled client refuses to build without an explicit
--user-confirmedflag and refuses quotes older than 60 seconds. - Numbers only from the API, with the chain, the source and the time named in every answer.
- Token safety. Canonical addresses from the curated list (and CoinGecko on Avalanche, Base, BNB Chain, Arbitrum and Ethereum), bridged versus native tokens (USDC vs USDC.e or USDbC), look-alike tickers, airdrop and approval scams, address poisoning, stock token versus share.
- Analysis is not advice: no buy calls, no promised returns, no stop-loss or take-profit recommendations.
| File | Contents |
|---|---|
SKILL.md |
when to use it, chain table, intent routing, confirmation flow, answer and safety rules, errors |
references/data.md |
market-data endpoints by chain |
references/quote-and-swap.md |
quotes, the unsigned transaction, approvals, simulation |
references/stocks.md |
Robinhood Chain stock tokens |
references/solana.md |
Solana: data and swaps |
references/ask.md |
Chord Ask over the API |
references/safety.md |
token and address safety |
references/examples.md |
real responses from the live API |
scripts/chord.py |
a dependency-free Python client (token, price, board — these three also on Solana — stock, quote, swap-tx, simulate, ask) |
scripts/install.sh |
the installer |
Other ways in
Section titled “Other ways in”The skill sits on the same public API as the MCP server, the function-calling manifest and the Docs MCP. If your agent already has the chord_* tools, the skill’s rules apply to them too: chord_swap_tx is the signing boundary.