The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill's primary functionality is to execute local Python scripts and shell commands on remote Google Colab virtual machines using the colab exec and colab run commands. This behavior is intentional and central to the skill's purpose for remote compute management.
[EXTERNAL_DOWNLOADS]: The skill recommends installing the google-colab-cli package via standard package managers (pip or uv). This tool is provided by the skill's author to interface with Google Colab services.
[COMMAND_EXECUTION]: The agent is instructed to run local shell commands including gcloud for authentication setup and colab for session management and execution.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection because it processes user-provided script files and command output without explicit boundary markers or sanitization.
Ingestion points: Local script files passed to colab exec -f or colab run, and piped stdin input.
Boundary markers: The instructions do not define delimiters to separate untrusted script content from agent instructions.
Capability inventory: Execution of arbitrary code on remote VMs, file system access (upload/download), and package installation (colab install).
Sanitization: There is no evidence of script content validation or sanitization prior to execution.

colab-operator