The Agent Skills Directory

[CREDENTIALS_UNSAFE] (HIGH): Exposure of cryptographic private keys in the repository file structure.
Evidence: The file structure in references/file_structure.md reveals files named 0x20c1DF6f3310600c8396111EB5182af9213828Dc.eth.pk.txt and BTV1zY7njS5an91v9nphCK48d2vnMuecEgHLYiP25ycj.sol.sk.json inside the tests/files/ directory. Storing raw private keys (Ethereum and Solana) in a repository, even for testing, is a critical security risk as they can be easily compromised.
[CREDENTIALS_UNSAFE] (MEDIUM): Insecure handling of wallet private keys via CLI parameters.
Evidence: The documentation in references/x402-turbo-upload.md provides examples where users are instructed to pass their private key directly as a command-line argument (e.g., --wallet "0xYOUR_PRIVATE_KEY"). Secrets passed as CLI arguments are logged in shell history (.bash_history) and are visible to other users/processes via system monitoring tools like ps.
[DATA_EXFILTRATION] (MEDIUM): Capability to read and transmit local files to external endpoints.
Evidence: The core purpose of the tool is to take a local file path (--path) and upload its contents to a remote service (https://upload.ardrive.io). While this is the intended functionality, an agent with access to this tool could be manipulated into exfiltrating sensitive system files (e.g., .env, SSH keys) if not strictly constrained.
[COMMAND_EXECUTION] (LOW): Execution of external scripts and installation of dependencies.
Evidence: The installation instructions in references/x402-turbo-upload.md require cloning a remote repository and running yarn and yarn build, which involves executing third-party code on the host machine.

turbo-sdk