The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes transaction and message signing requests received from external, potentially untrusted dApps via the WalletConnect protocol. This creates an attack surface where a malicious dApp could trigger the agent to sign transactions that result in the loss of funds.\n
Ingestion points: The session_request event listener in scripts/register-basename.js and scripts/wc-connect.js receives request parameters directly from the connected dApp.\n
Capability inventory: The skill utilizes ethers.js to sign messages (wallet.signMessage), sign typed data (wallet.signTypedData), and send transactions (wallet.sendTransaction).\n
Boundary markers: The documentation suggests using an --interactive mode to prompt for manual approval, but auto-approve is the default behavior.\n
Sanitization: No validation or sanitization of the transaction destination, value, or data is performed before signing and broadcasting.\n- [COMMAND_EXECUTION]: The scripts/register-basename.js script uses Puppeteer to launch and control a browser instance for automating dApp interactions.\n
Spawns browser: puppeteer.launch is called with --no-sandbox and --disable-setuid-sandbox arguments, which reduces process isolation.\n- [EXTERNAL_DOWNLOADS]: The skill depends on several external Node.js packages for core functionality.\n
Packages: @walletconnect/web3wallet, @walletconnect/core, ethers, and puppeteer are fetched from the npm registry.

walletconnect-agent