Soho

Initiate payments on the SOHO Pay credit layer using EIP-712 signatures.

This skill appears to implement exactly what it claims (EIP‑712 orchestration and on‑chain settlement) — but check a few things before installing/running: 1) Verify the registry metadata mismatch: skill.json / SKILL.md require RPC_URL, CHAIN_ID, and SIGNER_PROVIDER — make sure you supply and protect those env vars. 2) Only point WALLET_SIGNER_SERVICE_URL at a trusted, user‑controlled MPC/HSM/signing service and keep SIGNER_SERVICE_AUTH_TOKEN secret; audit that service’s /sign-eip712 and /send-tx endpoints. 3) Never use SOHO_DEV_PRIVATE_KEY on mainnet; the code enforces this but double-check CHAIN_ID and DEV_ALLOW_LOCAL_KEY before running. 4) Confirm the platform invocation settings — although the skill is written to be manual-invocation-only, the platform flag you were shown suggests autonomous invocation is permitted; ensure the platform honors manual-only invocation or that you don’t enable autonomous runs. 5) Note the declared SOHO_API_URL is not used by the code — ask the author if that’s intentional. 6) Because package.json exists, running locally requires npm install; review dependencies and the contract addresses in config before sending real funds. If anything above is unclear, ask the skill author for clarifications and an integrity checksum for the package files.

Type: OpenClaw Skill Name: soho-pay Version: 2.0.0 The OpenClaw AgentSkills bundle is designed for orchestrating payments on a credit layer and exhibits a strong security-first design. Key security features include explicit manual invocation and user confirmation requirements enforced by both metadata (`SKILL.md`, `skill.json`) and a runtime guard (`scripts/pay.js`). The skill strictly separates key custody, stating it never holds private keys, delegating signing to user-controlled wallet signers (local dev key or remote MPC/HSM). Robust environment variable validation using Zod (`src/config.js`) prevents misconfiguration, and the use of local private keys is strictly gated to testnets by default. Input sanitization for `merchantAddress` prevents address spoofing, and all external network calls are to user-configured URLs with proper authentication. There is no evidence of data exfiltration, malicious execution, persistence mechanisms, or prompt injection attempts against the agent.