The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill provides implementation examples that read private keys from specific local file paths, such as ~/.ethereum/keys/testnet.key and ~/.solana/keys/devnet.json. While intended for developer convenience, these instructions direct the agent to access sensitive credential files.\n- [EXTERNAL_DOWNLOADS]: The skill recommends installing several third-party Node.js packages for blockchain interactions, including viem, @solana/kit, @solana-program/token, ws, dotenv, and bs58.\n- [COMMAND_EXECUTION]: The skill instructs the user/agent to execute npm install commands to set up the necessary environment for USDC operations.\n- [INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it processes external inputs (addresses and amounts) within blockchain execution logic.\n
Ingestion points: User-provided transaction details in natural language prompts.\n
Boundary markers: The skill does not define specific structural boundaries or delimiters for user-provided data.\n
Capability inventory: Sensitive file read access (fs.readFileSync) and blockchain state-modifying write operations (transfer, approve).\n
Sanitization: The instructions emphasize manual validation and user confirmation but lack automated sanitization or escaping mechanisms for the interpolated strings.

use-usdc