The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user content (topics, styles, and character identity anchors) and integrates it directly into prompts for downstream AI models like Seedance and Gemini. This creates a vulnerability where a malicious user could supply inputs designed to override the behavior of subsequent models in the processing chain.\n- Ingestion points: Input variables with.task, with.request, with.render_style, and with.identity_anchor as defined in SKILL.md.\n- Boundary markers: The skill does not utilize specific delimiters (e.g., XML tags or clear 'end-of-user-content' markers) to segregate user-supplied values from the generated instructions.\n- Capability inventory: The skill itself is restricted to text generation and lacks the capability to execute shell commands, perform network requests, or access sensitive files.\n- Sanitization: Rule 9 requires plain text output to prevent encoding failures (UnicodeEncodeError) in downstream processing environments, and Rule 3 provides negative constraints regarding the use of protected intellectual property.

ai-video-script