Monolith — Crypto Wallet

Secure crypto wallet for AI agents. Hardware-isolated keys (Apple Secure Enclave), ERC-4337 smart wallet, on-chain spending caps, default-deny policy engine.

This skill appears to do what it says: it builds transaction intents and communicates with a local macOS daemon that performs signing and enforces policy. Before installing: 1) Treat the MonolithDaemon.pkg as a privileged install (requires admin) — verify the release via checksums/signatures and the GitHub project lineage. 2) Inspect or vet the daemon binary/source (the JS here talks only to the daemon; the daemon actually holds keys and does signing). 3) Limit exposure: use small balances and strict per-tx/daily caps and an allowlist before giving the agent autonomous invocation rights. 4) Note the registry vs SKILL.md inconsistency: manifest says instruction-only but the skill includes code and install entries — confirm you understand the full install flow. 5) If you will allow autonomous agent actions, prefer an interactive approval path (do not run headless) and keep tight policy settings. If you want, provide the daemon binary hash or a link to a signed release and I can point out what additional checks to perform.

Type: OpenClaw Skill Name: monolith Version: 0.1.10 The skill implements a robust security model where the 'untrusted' skill only builds transaction intents, and a separate, trusted, local macOS daemon enforces policy and handles signing with human approval (Touch ID). The code in `lib/intent-builder.js` and `scripts/*.js` adheres to this by communicating with the daemon via a Unix socket (`lib/daemon-client.js`) and making read-only network calls to legitimate public RPCs and the Uniswap API. Crucially, `lib/runtime-bootstrap.js` explicitly avoids automatic execution of system commands for installation or persistence, instead providing manual instructions. There is no evidence of data exfiltration, unauthorized execution, persistence mechanisms, or prompt injection attempts against the agent.