The Agent Skills Directory

[PERSISTENCE]: The installation script located at references/install.sh modifies the user's shell configuration files (~/.bashrc and ~/.zshrc) to add the installation directory to the PATH environment variable. This ensures the qshell command remains accessible in subsequent terminal sessions.
[REMOTE_CODE_EXECUTION]: The skill automates the download and execution of the qshell binary from kodo-toolbox-new.qiniu.com. This installation process is triggered if the CLI tool is not detected on the system, retrieving software directly from the service provider's infrastructure.
[CREDENTIALS_UNSAFE]: The skill facilitates the configuration of AccessKey and SecretKey through the qshell account command. It mandates a security best practice by instructing the agent to prefix the command with a leading space to prevent credentials from being recorded in the shell's command history.
[COMMAND_EXECUTION]: The skill is primarily a wrapper for the qshell command-line utility, allowing the agent to perform a wide variety of storage management tasks, including file operations and sandbox execution.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect injection as it ingests and displays external data such as bucket file lists and resource metadata.
Ingestion points: Data enters the system context via qshell listbucket2, qshell stat, and similar listing commands.
Boundary markers: No explicit delimiters or instructions are used to distinguish untrusted data from system prompts.
Capability inventory: The skill can perform network operations, file modifications, and command execution within sandboxes.
Sanitization: There is no evidence of specific sanitization or filtering of the metadata retrieved from the object storage service.

qshell