pocket-money

Give your AI agent crypto wallets on Base. Create purpose-specific wallets, ask your human to fund each one, check balances, and manage budgets. No accounts,...

This skill is coherent with its stated aim (creating local Base/USDC wallets), but it has security-sensitive components you should review before installing: 1) Inspect the @auteng/pocket-money package source on GitHub and the npm package contents (including dependencies and postinstall scripts). 2) Do not fund these wallets with more than a small, reviewed amount — treat them as petty cash. 3) Consider requiring encrypted keystores or a hardware signing flow rather than unencrypted JSON files on disk; if you must use these files, keep strict filesystem permissions and backups off the machine. 4) Confirm the RPC endpoints (mainnet.base.org) and whether gas in ETH is actually required — the SKILL.md claim that "no ETH needed for gas" may be inaccurate on Base unless a gas-sponsorship mechanism is used. 5) If possible, run the npm package in a sandboxed environment or audit it first; pin to a specific package version and vendor. 6) Be aware that the agent can be invoked autonomously; if you want to prevent accidental spending, restrict autonomous invocation for this skill or ensure that human approval can be enforced by policy. If you cannot audit the package and accept these risks, proceed cautiously; otherwise, treat this skill as potentially dangerous.

Type: OpenClaw Skill Name: pocket-money Version: 2.0.0 The skill stores unencrypted private keys for crypto wallets on disk at `.auteng/wallets/<name>.json`, which is a significant vulnerability as it exposes funds if the host machine is compromised. While the `SKILL.md` transparently discloses this risk and provides mitigations (e.g., restricted file permissions, advice to use small amounts, and explicit instructions for the AI agent to seek human approval before any spending), the inherent risk of unencrypted key storage makes it suspicious. There is no evidence of intentional malicious behavior like data exfiltration or unauthorized remote control.