fiftyone-find-duplicates
Fail
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: CRITICALREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The README.md file recommends installing the skill using 'curl -sL skil.sh | sh', which pipes a remote script from an unknown domain directly into the shell for execution. This practice bypasses security reviews and allows for arbitrary code execution from a source that is not recognized as a trusted vendor.
- [EXTERNAL_DOWNLOADS]: The skill fetches code from 'skil.sh' and also downloads plugins from the 'voxel51/fiftyone-plugins' repository on GitHub. While the GitHub repository belongs to the vendor, the 'skil.sh' domain is an untrusted external source.
- [INDIRECT_PROMPT_INJECTION]:
- Ingestion points: The skill ingests and processes images and metadata from FiftyOne datasets as described in SKILL.md.
- Boundary markers: None are present; the instructions do not include delimiters or warnings to ignore instructions embedded in dataset content.
- Capability inventory: The skill uses high-privilege tools including 'execute_operator', 'download_plugin', 'launch_app', and 'set_view'.
- Sanitization: There is no evidence of sanitization or validation of the dataset content before it is processed by brain similarity operators.
- [COMMAND_EXECUTION]: The troubleshooting section instructions include the execution of shell commands such as 'pip install torch' to manage dependencies, which could be exploited if malicious requirements are introduced.
Recommendations
- AI detected serious security threats
Audit Metadata