The Agent Skills Directory

[SAFE]: The skill consists of instructional markdown files that guide the AI in generating video prompts. All logic is textual and does not perform any unauthorized system operations or malicious behaviors. \n- [INDIRECT_PROMPT_INJECTION]: The skill has an attack surface where it ingests untrusted user ideas and images (Step 0 in SKILL.md). It mitigates this via a mandatory evidence chain: 1. Ingestion point is defined in Step 0; 2. Boundary markers are enforced through a strict four-block output schema (references/output-format.md); 3. The skill's capabilities are limited to text generation with no file-system or network write access; 4. A specific safety and desensitization step (Step 7 in SKILL.md) identifies and rewrites sensitive or dangerous content into safe alternatives. \n- [EXTERNAL_DOWNLOADS]: The README.md documentation references the author's GitHub repository and an installation CLI tool. These are standard methods for distributing agent skills and do not involve runtime network calls by the skill itself during execution. \n- [COMMAND_EXECUTION]: The installation guide in README.md mentions shell commands like npx, git, and python. These are instructions for the human user to set up the tool manually and are not triggered or executed by the AI agent as part of the skill's operational runtime.

seedance-director