The Agent Skills Directory

[DYNAMIC_EXECUTION]: The agent is instructed to use eval() to load the scripts/publication_batch.js helper script into its execution environment. This provides the helper script with direct access to the agent's toolset.
[DYNAMIC_EXECUTION]: The script scripts/speaker_effects.py executes JavaScript code extracted from the user-controlled project file (index.html) by spawning a Node.js subprocess. Although it attempts to sandbox this execution using the vm module and by disabling certain global functions, the Node.js vm module is not designed to be a secure sandbox for untrusted code.
[INDIRECT_PROMPT_INJECTION]: The skill ingests 'notes' from web assignments and reference materials, treating them as editorial instructions. This creates a surface where malicious instructions embedded in external data could influence the agent's actions during the video production process.
[COMMAND_EXECUTION]: The skill relies on executing several local Python and JavaScript utility scripts to manage the video editing workspace, synchronize files, and perform network uploads. These scripts have broad access to the local project files and network capabilities.

aip