The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines complex multi-step workflows where output from one AI model (e.g., a script from Claude) is used as input for another model (e.g., Kokoro TTS or OmniHuman). This creates a surface for indirect prompt injection where instructions could be embedded in intermediate data to influence subsequent steps.
Ingestion points: The pipeline scripts ingest text outputs, image URLs, and audio URLs across different bash steps (e.g., script.json to voice.json).
Boundary markers: Absent. The skill uses standard JSON input formats without specific delimiters or instructions to ignore embedded commands in the processed data.
Capability inventory: The skill has the ability to execute belt CLI commands through Bash to perform network-based media generation and merging.
Sanitization: There is no evidence of sanitization or filtering applied to the intermediate data passed between pipeline steps.
[EXTERNAL_DOWNLOADS]: The skill references and downloads documentation and configuration from inference.sh and its associated GitHub repository (github.com/inference-sh). These downloads are integral to the skill's stated purpose of using the platform's CLI.
[COMMAND_EXECUTION]: The skill utilizes the belt CLI via Bash to perform its primary functions. The execution is constrained to the belt binary as specified in the skill's allowed-tools configuration.

ai-content-pipeline