Onchain CLI

CLI for crypto portfolio tracking, market data, CEX history, and transaction lookups. Use when the user asks about crypto prices, wallet balances, portfolio values, Coinbase/Binance holdings, Polymarket predictions, or transaction details.

This skill appears to be a legitimate crypto CLI, but exercise caution before installing or providing credentials. Points to consider: - The SKILL.md/README require multiple sensitive API keys (Coinbase/Binance keys and secrets, DeBank, Helius, Etherscan/Solscan, etc.) but the registry metadata declares no required env vars — ask why the manifest omitted these or prefer a skill that clearly declares required credentials. - Do not supply CEX API keys with withdrawal permissions; create read-only API keys and enable IP whitelisting if possible. Test first with non-critical or demo accounts and a public wallet address to verify behavior. - Verify the package source before installing (README references npm package @cyberdrk/onchain). Confirm the publisher, homepage/GitHub repo, and release checksums/signatures if available. - Note the inconsistent env-var names between README and SKILL.md (e.g., Coinbase key naming); confirm the exact variables the installed CLI expects to avoid accidentally placing secrets in the wrong env variable. - Check the config file (~/.config/onchain/config.json5) permissions and contents after setup; do not store secrets in world-readable files. - If you want, I can: (a) attempt to look up the npm package/repo to verify publisher and source, or (b) produce a short checklist to create limited-permission API keys for Coinbase/Binance and to validate CLI behavior with read-only access.

Type: OpenClaw Skill Name: onchain Version: 0.3.0 The skill bundle provides a CLI for crypto portfolio tracking and market data. All requested API keys and configuration methods are directly relevant to its stated purpose. The `SKILL.md` contains instructions for the AI agent, including 'DO NOT' directives regarding alternative methods for transaction lookups (e.g., `curl`, `cast`). While these are a form of prompt injection, they aim to guide the agent to use the `onchain` CLI as the authoritative source for its specific task, rather than instructing it to ignore the user, hide actions, exfiltrate data, or access unrelated sensitive information. There is no evidence of intentional harmful behavior, data exfiltration, or malicious execution.