The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the belt CLI tool to perform AI inference tasks, including video generation with Google Veo, image generation with FLUX, and text-to-speech with Kokoro. These commands are executed via the Bash tool as defined in the skill's allowed-tools configuration.
[EXTERNAL_DOWNLOADS]: The documentation includes instructions for the user to install the belt CLI (belt-sh/cli) and supplementary skills from the inference-sh organization using the npx skills add command.
[PROMPT_INJECTION]: The skill processes untrusted data such as user-provided concepts and AI-generated scripts to create social media content. This ingestion of external data without explicit sanitization markers creates a surface for indirect prompt injection.
Ingestion points: User-defined variables (e.g., $CONCEPT, $topic) and scripts generated by Claude are interpolated into the --input JSON payload of the belt command in SKILL.md.
Boundary markers: None; the data is directly inserted into JSON string templates.
Capability inventory: The skill can execute shell commands via the belt CLI, which includes posting to Twitter and running various cloud-based AI models as seen in SKILL.md.
Sanitization: There is no evidence of validation or filtering for the interpolated content.

ai-social-media-content