The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided TikTok scripts and captions, which serves as a surface for indirect prompt injection. Malicious instructions could be embedded within the script to manipulate the agent, especially given its capabilities to interact with external tools and publishing layers.
Ingestion points: Untrusted scripts and captions are ingested via the tt-humanizer <script> input defined in SKILL.md.
Boundary markers: The skill lacks explicit delimiters or instructions for the agent to ignore any command-like text within the user scripts.
Capability inventory: The agent has access to image generation and refinement via lib.illustrate and refine in sub-skills/illustration.md, and mentions a 'Publora' publishing layer.
Sanitization: There is no documentation of sanitization or filtering of the script content before it is processed by the agent's logic.
[DYNAMIC_EXECUTION]: The skill contains Python code snippets in references/scrub-rules.md and sub-skills/illustration.md that define core logic for script scoring and tool invocation. These snippets represent a dynamic execution pattern where the agent follows structured code templates to perform its tasks, such as regex-based text analysis and image generation through internal libraries.

tt-humanizer