The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes platform-specific CLI tools (infsh app run and belt task cost) to orchestrate generative AI services for image, audio, and video production. This usage is well-defined and aligns with the primary purpose of the skill.
[INDIRECT_PROMPT_INJECTION]: The pipeline ingests user-provided text scripts and image URLs into generation tools. This represents a typical attack surface for indirect prompt injection where instructions could be embedded in the input data. However, the skill uses structured JSON inputs, providing inherent boundary markers between instructions and data.
Ingestion points: User-supplied strings for phota/train (image URLs), inworld/text-to-speech-2 (text turns), and pruna/p-video-avatar (video prompts) in SKILL.md.
Boundary markers: Commands utilize JSON formatting for inputs, which helps isolate data from tool execution parameters.
Capability inventory: Uses infsh app run for service interaction and Write for saving output JSON files as specified in SKILL.md.
Sanitization: None explicitly defined within the skill instructions; it relies on platform-level tool sanitization.
[SAFE]: No obfuscation, data exfiltration, or unauthorized persistence mechanisms were identified. All external service calls are directed to recognized platform components and follow established usage patterns for the intended use case.

ai-podcast