The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface where external or untrusted data is interpolated into AI model prompts.
Ingestion points: SKILL.md (e.g., 'Blog to Video Pipeline' and 'YouTube Short Pipeline' sections).
Boundary markers: Absent. Untrusted inputs such as <blog-content>, <script-text>, and <point-text> are placed directly into JSON input payloads for models like Claude without delimiters or instructions to ignore embedded commands.
Capability inventory: The skill employs the Bash tool to execute belt app run commands, which interact with remote AI services and write outputs to local JSON files.
Sanitization: Absent. The skill does not demonstrate escaping or validation of the content before it is processed by the AI models.
[EXTERNAL_DOWNLOADS]: The skill references and fetches guidelines and installation instructions from the official repository of the inference-sh platform.
Evidence: Downloads from https://raw.githubusercontent.com/inference-sh/skills/refs/heads/main/cli-install.md and refers to npx skills add commands for platform-specific extensions.

ai-content-pipeline