The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's utility script srt_utils.py executes several local commands to identify system capabilities and optimize performance.
Evidence includes list-based subprocess calls to nvidia-smi, sysctl, wmic, and fc-list to detect GPUs, memory, and installed fonts.
These diagnostic commands do not use a shell and do not incorporate unvalidated user input, making them relatively safe.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of YouTube video titles and transcription text.
Ingestion points: Subtitle files are parsed in srt_utils.py, and transcripts are passed to the agent for translation in Step 3 of the pipeline.
Capability inventory: The skill has the ability to execute media tools via bash and read/write files in the local environment.
Boundary markers: While Step 3 provides strict formatting rules for translation, it lacks explicit 'ignore embedded instructions' markers or delimiters to protect against adversarial content inside the transcript.
Sanitization: Basic normalization is performed on slugs, but transcription text is processed as raw strings.
[DYNAMIC_EXECUTION]: The script uses dynamic importing to handle an optional dependency.
srt_utils.py uses importlib.import_module('unidecode') to transliterate non-Latin characters if the package is present on the system. This is a standard practice for handling optional features but is noted as dynamic code loading.

yt2bb