The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The installation instructions in README.md direct users to execute a script directly from a remote URL ('curl -sL skil.sh | sh'). This pattern bypasses package management and security audits, allowing for arbitrary code execution from a third-party domain.
[PERSISTENCE]: The skill provides commands to persistently modify the user's shell environment by appending settings to profile files such as '~~/.zshrc', '~~/.bashrc', and '~/.bash_profile'. This is used to set ulimit values and environment variables like FIFTYONE_DATABASE_NAME.
[COMMAND_EXECUTION]: The skill includes powerful and potentially destructive shell commands, such as 'rm -rf ~/.fiftyone/var/' for resetting the MongoDB database and 'pkill -f "fiftyone"' for terminating processes. While these are gated by user confirmation directives, they represent a significant capability surface.
[EXTERNAL_DOWNLOADS]: The skill uses 'fo.download_plugin' to fetch code from 'voxel51/fiftyone-plugins' on GitHub. As the skill is authored by voxel51, this is an expected vendor resource, though it involves fetching and executing external code.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided symptoms to trigger specific troubleshooting workflows that involve shell command execution and file system modifications. 1. Ingestion points: User symptoms/queries provided to the assistant (referenced in README.md). 2. Boundary markers: None identified for user-provided data. 3. Capability inventory: File system writes ('echo >>'), destructive deletions ('rm -rf'), process termination ('pkill'), and package installation ('pip install'). 4. Sanitization: No explicit sanitization or validation of the user's input before processing is described.

fiftyone-troubleshoot