The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to perform operations on remote servers via SSH. To mitigate security risks, it provides utility scripts (remote-cmd and remote-bash) that use shell-safe quoting (via printf '%q'). This ensures that arguments are treated as literals by the remote shell, preventing local shell expansion of variables or command substitution from affecting the remote execution environment.
[REMOTE_CODE_EXECUTION]: Through the remote-bash --upload functionality, the skill can transmit and run local scripts on a remote host. The implementation uses secure piping and quoting to prevent injection during the process. Furthermore, the skill's instructions prioritize the use of pre-existing server-side wrappers to limit the frequency and risks associated with remote script execution.
[SAFE]: The skill emphasizes secure data management by instructing the agent to store user-specific credentials, SSH aliases, and private paths in a local override file (.agent/local-overrides.yaml) that is gitignored. This prevents the accidental exposure of sensitive environment details in shared repositories.
[SAFE]: The instructions provide robust logic for the agent to handle network access issues in sandboxed environments. It specifically teaches the agent to differentiate between network reachability failures and actual authentication errors (e.g., for gh or git), reducing the likelihood of misinterpreting security state during troubleshooting.

remote-project-control