The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides an installation command that fetches and executes a script from a remote server using a shell pipe: bash -c "$(curl -fsSL http://xrl.alibaba-inc.com/install.sh)". Additionally, the CLI tool is described as having an automatic update feature for minor versions. The source domain belongs to a well-known technology organization.\n- [COMMAND_EXECUTION]: The CLI includes functionality to execute arbitrary commands within a sandbox environment using the exec and execute subcommands (e.g., rockcli sandbox <id> exec '<command>') as described in references/sandbox.md.\n- [DATA_EXFILTRATION]: The skill includes tools for bidirectional file transfer between the local system and the sandbox environment using upload and download commands. It also provides access to sandbox history and logs through the history and log commands.\n- [PROMPT_INJECTION]: The agent run feature processes local configuration files (job.yaml) and project directories (upload --dir). This presents an indirect prompt injection surface if the processed data originates from untrusted sources.\n
Ingestion points: agent run -c job.yaml (in references/agent.md) and upload --dir (in SKILL.md).\n
Boundary markers: None specified in the provided instructions to isolate or ignore instructions within these files.\n
Capability inventory: The skill allows for arbitrary command execution (exec), interactive sessions (attach), and file system operations (upload/download) as defined in references/sandbox.md.\n
Sanitization: No sanitization or validation logic is mentioned for the content of the configuration or project files.

rock-cli