Venus Agent Skills

Analyze Venus Protocol lending/borrowing positions on BNB Chain with risk-first guidance. Use when users ask about Venus markets, collateral/borrow decisions...

This package appears to do what it claims: market reads, on‑chain wallet exposure checks, conservative risk scoring, and optional transaction helpers. Important things to consider before installing or running: 1) Default mode is simulate/read-only — always run simulate/preview first. 2) Broadcast actions require you to supply a private key and --confirm YES; avoid passing private keys on the command line (they can appear in process lists and shell history). Prefer using a secure signer (hardware wallet, ephemeral node signer) or a protected key file. 3) Verify RPC and API endpoints before use (defaults are public BNB dataseed and api.venus.io); do not point the scripts to an untrusted RPC or API. 4) Review/verify contract addresses in references (pool-filter.json and references JSON) against official sources. 5) If you intend to let an agent invoke this autonomously, restrict that capability or require explicit human confirmation for any broadcast step. If you want a deeper check, request an audit of the transaction broadcast code paths (node scripts that call wallet.signer) and of wallet_onchain_exposure.py to confirm no unexpected network calls or data exfiltration.

Type: OpenClaw Skill Name: venus-agent-skills Version: 1.0.0 The skill bundle is designed for risk-aware analysis and operations on Venus Protocol and Flux. It includes clear guardrails in `SKILL.md` and the scripts, such as defaulting to 'simulate' mode, requiring explicit '--confirm YES' and '--private-key' for 'broadcast' transactions, and implementing health factor safety checks (e.g., in `scripts/venus_withdraw.js`). While scripts use `subprocess.check_output` (Python) and `execFileSync` (Node.js) to call other scripts, arguments are passed as lists, which is the standard and generally safe method to prevent shell injection. There is no evidence of intentional data exfiltration, unauthorized persistence, obfuscation, or malicious prompt injection attempts against the agent.