The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's operational logic depends on executing a sequence of shell commands to interact with specialized AI media tools.\n
Evidence: The workflow utilizes infsh app run to invoke multiple services for character generation, voice synthesis, and video stitching.\n
Evidence: It uses the belt CLI tool to perform cost monitoring of the ongoing generation tasks.\n- [INDIRECT_PROMPT_INJECTION]: The skill accepts and processes untrusted data from external sources, which could lead to instructions being executed by the agent if they are embedded within the inputs.\n
Ingestion points: The skill ingests user-provided conversation scripts for TTS generation and external image URLs for training character identities in the phota/train step.\n
Boundary markers: Data is passed as JSON parameters to the tools, but the instructions do not include explicit delimiter patterns or system-level directives for the agent to ignore potentially adversarial content inside the script or image metadata.\n
Capability inventory: The skill has access to powerful tools including shell execution (Bash), persistent storage interaction (Read, Write), and recursive tasking (Agent).\n
Sanitization: No sanitization, escaping, or verification steps are defined for the input strings before they are interpolated into the execution pipeline.

ai-podcast