The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from video transcripts and library item metadata (captions, URIs) which is used to generate video editor commands. There is a theoretical surface for malicious content in a transcript to influence the agent's behavior during the specification generation phase.\n
Ingestion points: Video transcripts via get_transcript and user-saved items via list_library.\n
Boundary markers: No specific delimiters or safety instructions to ignore embedded commands are present in the instruction steps.\n
Capability inventory: The skill executes local Python scripts via python3 and utilizes the apply_freeform_editor_commands tool.\n
Sanitization: No explicit sanitization or escaping of transcript text is implemented in the provided scripts.\n- [EXTERNAL_DOWNLOADS]: The common.py script downloads a font from the official Google Fonts GitHub repository. This is a trusted source used for legitimate layout calculations and does not represent a security risk.\n- [COMMAND_EXECUTION]: The skill utilizes command-line execution to run its internal Python scripts (statement.py, inset.py) for processing animation specifications.

minimal-statements