The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/generate_variants.py executes ffmpeg using subprocess.run to apply video effects. The command arguments, including video filters and speed settings, are retrieved from hardcoded configuration maps within the script, which mitigates the risk of arbitrary command injection.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it reads content from a user-provided JSON file and interpolates it directly into an LLM prompt.
Ingestion points: The scripts/generate_variants.py script reads the content of a local file specified by the --base-copy argument.
Boundary markers: Absent. The JSON content is dumped directly into the prompt string without delimiters or instructions to ignore embedded commands.
Capability inventory: The skill has the ability to execute ffmpeg commands via subprocess and perform network requests to an external API (Volcengine Ark).
Sanitization: None. The script uses json.dumps but does not sanitize or validate the content for instructions that might subvert the LLM's behavior.

ad-abtest