The Agent Skills Directory

[COMMAND_EXECUTION]: The run_command tool in tools.ts and handler.ts allows the execution of arbitrary shell commands within the project workspace.\n- [DYNAMIC_EXECUTION]: The execute_code tool enables dynamic execution of Python and Node.js snippets at runtime, utilizing a CodeSandbox for execution.\n- [REMOTE_CODE_EXECUTION]: The run_ssh_command tool provides a mechanism to execute commands on remote systems via an OpenSSH client.\n- [EXTERNAL_DOWNLOADS]: The install_package tool allows the agent to download and install packages from external npm and pip registries, introducing potential supply chain risks.\n- [DATA_EXFILTRATION]: The ssh_copy_file tool supports a push direction that enables transferring files from the local workspace to a remote host, facilitating data exfiltration.\n- [INDIRECT_PROMPT_INJECTION]: The skill exposes a significant attack surface for instructions embedded in untrusted data that the agent may process.\n
Ingestion points: The code and command arguments of the execute_code, run_command, and run_ssh_command tools in handler.ts.\n
Boundary markers: None; the handler processes and executes the provided strings without using delimiters or safety-oriented boundary markers.\n
Capability inventory: The skill has the ability to execute local and remote commands, run dynamic code, install external dependencies, and perform network file transfers.\n
Sanitization: No sanitization or safety checks are performed on the input strings before they are passed to the execution environment.

code-execution