The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection as it processes untrusted media content and prompts using external tools.
Ingestion points: Media files (images, videos, audio) and text prompts are passed to the bl CLI through flags like --image, --video, --audio, --url, and --prompt across all command groups (bl omni, bl vision, bl speech, bl video).
Boundary markers: The skill does not provide instructions to the agent to use delimiters or ignore potential commands hidden within the media content or metadata.
Capability inventory: The skill uses the bl CLI for network API interaction and file management, and utilizes ffmpeg for file concatenation and audio mixing, which involves reading and writing local files.
Sanitization: No input sanitization or content validation is performed on the media files before they are processed by the CLI utilities.
[COMMAND_EXECUTION]: The skill executes the bl command-line tool and potentially the ffmpeg utility. These tools have permissions to interact with the network and access the local file system to perform the skill's media generation and transformation tasks.

bailian-gen