The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions frequently utilize the Bash tool to run the belt CLI. This involves executing commands like belt app run to interact with external AI services (FLUX, Wan, Kokoro, etc.), which is the primary intended function of the skill.
[INDIRECT_PROMPT_INJECTION]: The skill establishes a pipeline where user-provided text (prompts, scripts, and blog content) is interpolated into CLI commands and processed by external AI models.
Ingestion points: User-supplied strings are ingested into the prompt and input fields of the belt app run commands as demonstrated in SKILL.md examples.
Boundary markers: The skill uses JSON structure (--input '{...}') to delimit inputs, which provides some structural separation but does not prevent malicious instructions within the strings themselves.
Capability inventory: The skill uses the Bash(belt *) tool to execute shell commands, perform media merging via infsh/media-merger, and handle file redirects (e.g., > script.json).
Sanitization: There is no evidence of explicit input sanitization or validation within the skill instructions; it relies on the downstream AI models and the CLI tool's own handling.

ai-content-pipeline