fiftyone-dataset-import
Warn
Audited by Gen Agent Trust Hub on Jun 16, 2026
Risk Level: MEDIUMREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill directs the agent to use pickle.load() on annotation files found within datasets. Unsafe deserialization of untrusted data using the pickle module is a vulnerability that can lead to arbitrary code execution.
- [REMOTE_CODE_EXECUTION]: The skill implements a dynamic package discovery workflow that suggests searching for and installing Python packages from PyPI and GitHub based on folder contents. This automated installation of remote code from unverified sources increases the risk of executing malicious software or typosquatted dependencies.
- [EXTERNAL_DOWNLOADS]: The skill performs downloads from external sources, including GitHub repositories and the Hugging Face Hub, to obtain dataset processing tools and data files.
- [COMMAND_EXECUTION]: The skill uses multiple shell commands, including find, ls, git, and pip, to scan directories and manage dependencies. These commands operate on data and paths discovered at runtime, creating an attack surface for command-related issues.
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted external data, such as file names, directory structures, and Hugging Face repository READMEs. This data enters the agent's context without sanitization or boundary markers, allowing for potential manipulation of agent behavior during the scan and import process.
Audit Metadata