The Agent Skills Directory

[SAFE]: The skill is authored by MetaMask and provides documentation and workflows for the official MetaMask Agentic CLI.
[COMMAND_EXECUTION]: The skill uses the mm CLI for all blockchain operations. Write operations such as transfers and signing require explicit user confirmation and are subject to server-enforced MFA and wallet policies.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of the @metamask/agent-wallet CLI from npm and manages third-party plugins. These are standard features of the platform and are accompanied by explicit warnings regarding the trust model for third-party code.
[DYNAMIC_EXECUTION]: The scripts/x402_pay.py helper script uses subprocess.run to delegate transaction signing to the mm CLI. The script includes input validation for addresses and amounts, and enforces HTTPS for remote resources to mitigate security risks. The use of subprocess is restricted to calling the trusted mm binary with specific, validated arguments.
[CREDENTIALS_SAFE]: The skill follows security best practices by instructing users to store mnemonics and passwords in environment variables (MM_MNEMONIC, MM_PASSWORD) rather than passing them as plain-text CLI flags, preventing exposure in shell history.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data from blockchain APIs and x402 payment challenges. However, it incorporates strong defensive instructions requiring the agent to show decoded intents to the user and obtain explicit manual approval for every transaction, effectively mitigating the risk of autonomous unauthorized actions.

metamask-agent-wallet