openfinance-tech/skills

Agent skills for the OpenFinance backend — curated playbooks that teach AI agents how to trade on Polymarket and Hyperliquid, bridge via Relay, and use the user's OpenFinance-managed wallets correctly.

Skills lead with the actual backend routes (/agent/…) and WebSocket channels, so they're useful whether you're integrating via direct HTTP/WS, through an SDK, or via the OpenFinance MCP.

Install

# Project-level (recommended — commit with your codebase)
npx skills add openfinance-tech/skills

# Or global
npx skills add openfinance-tech/skills -g

# Install to a specific agent only
npx skills add openfinance-tech/skills -a claude-code

Requires the CLI from vercel-labs/skills.

Skills included

Skill	Triggers on
openfin-setup	First-time user, API key check, 401/412 auth errors
openfin-troubleshooting	"Why is this failing", allowance errors, RPC issues, setup-incomplete errors
openfin-polymarket	Markets, orderbooks, orders, positions/PnL, leaderboard, deposit / withdraw via bridge.polymarket.com
openfin-hyperliquid	Perp/spot trading, leverage, TWAP, WS market data, unifiedAccount auto-upgrade
openfin-relay	Cross-chain swaps, bridging, Solana routes, bridge+call
openfin-onchain	Token metadata, wallet portfolios, balances, USD prices, same-chain transfers
openfin-onramp	Fiat → crypto via Moonpay (cards, global) or Onramp.money (UPI / IMPS, India)

Backend prerequisites

Skills assume a running OpenFinance backend with embedded wallets provisioned for each user. See openfinance-tech docs for deployment and environment setup.

Risk & audits

These skills sign real on-chain transactions on a user's wallet through the OpenFinance backend. Registry auditors flag the trading skills as MEDIUM-risk on that basis, and that flag is correct:

Auditor	Finding	What it means
Snyk W009	"Direct money access capability detected" (risk 1.00)	The skill drives the OpenFinance backend's signing endpoints. Category-based rule — fires on any signing skill.
Socket	SUSPICIOUS MEDIUM	"Purpose-aligned, no install-chain or malware indicators" but enables real-world asset movement.

Neither finding identifies a malicious pattern, dependency hazard, or hidden behavior. Both are honest signals that the skills can move funds, which is the whole point.

What's already in place to keep that capability safe:

• Each transactional skill ships an explicit ## Safety contract — read-only quote first, full disclosure of amounts/chains/fees, explicit per-write user confirmation in chat, no recipient/contract addresses pulled from untrusted content, re-quote on any parameter change. See SECURITY.md for the repo-wide version.
• External recipients require an explicit verbatim warning. The caller's own wallets (EVM EOA + Solana via get_wallet_addresses, Polymarket deposit wallet via get_deposit_wallet) are resolved before any send/bridge/withdraw write; if the destination isn't one of those, the agent surfaces a bold "⚠️ EXTERNAL TRANSFER — funds cannot be recovered if the address is wrong. Type 'yes' to confirm." and proceeds only on explicit "yes" in the same turn.
• No secrets, scripts, or binaries in this repo — only markdown. Network calls happen from the agent and the OpenFinance backend, not from anything installed by npx skills add.

Found a way these skills could mislead an agent into a write without confirmation? Open an issue.