The Agent Skills Directory

[COMMAND_EXECUTION]: Documentation describes the vm.ffi cheatcode, which enables execution of arbitrary shell commands and external binaries from within Solidity test environments. Evidence: Detailed in references/features-ffi-signatures.md.
[CREDENTIALS_UNSAFE]: Instructions guide the user on providing sensitive data such as private keys for transaction signing (--private-key) and Etherscan API keys for contract verification. Evidence: Mentioned in references/features-cast.md and references/features-coverage-verify.md.
[EXTERNAL_DOWNLOADS]: The skill references methods for retrieving external code, including the forge install command for managing dependencies via git and the use of npx to run the Chisel REPL. Evidence: Documented in references/core-project-layout.md and references/features-chisel.md.
[REMOTE_CODE_EXECUTION]: The skill documents an environment susceptible to indirect prompt injection where untrusted Solidity code could trigger dangerous capabilities. Ingestion points: The agent is directed to work with Solidity files in src/, test/, and script/ directories, and install external libraries via forge install (as seen in core-project-layout.md). Boundary markers: There are no documented boundary markers or instructions to treat external Solidity code as untrusted data to prevent unintended command execution. Capability inventory: The documented toolset includes the ability to execute arbitrary shell commands via vm.ffi (features-ffi-signatures.md) and perform network operations using private keys (features-cast.md). Sanitization: While the documentation mentions that FFI must be explicitly enabled, it does not provide strategies for an agent to validate or sanitize Solidity inputs before execution.

foundry