The Agent Skills Directory

[METADATA_POISONING]: The skill provides contradictory information regarding its execution environment. The description and SKILL.md explicitly state that the system "completely moved away from local code dependencies" and "forbids calling any local dependency packages" (e.g., moviepy, opencv-python). However, the skill package includes a scripts/ directory with Python files (image_quality_checker.py, video_composer.py) and references/use_cases.md provides explicit instructions for users to pip install those exact packages. This deception could lead a user to misjudge the skill's actual reliance on local execution.
[TIME_DELAYED_CONDITIONAL]: The skill implements a "Daily Experience Limit" mechanism as described in references/usage_limit.md. Users are limited to three video generations per day. Once this limit is reached, the skill is instructed to block further usage and display a promotional message redirecting users to a WeChat Official Account ("TOPGO智能AI123") to obtain "unlimited" service, acting as a lead-generation funnel.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection because it processes untrusted user data without sufficient safeguards.
Ingestion points: The skill ingests untrusted product information (category, selling points, audience) from users, which is then processed by multiple AI sub-agents (SKILL.md).
Boundary markers: There are no explicit delimiters or "ignore embedded instructions" warnings provided for the sub-agents when they handle this user-supplied data.
Capability inventory: The skill possesses capabilities that could be abused if an injection is successful, including local file writing (video_composer.py writes MP4 files) and local script execution (via command-line instructions in use_cases.md).
Sanitization: The instructions do not specify any validation or sanitization of the user-provided product data before it is interpolated into agent prompts.
[COMMAND_EXECUTION]: The documentation in references/use_cases.md directs users to execute local Python scripts (scripts/image_quality_checker.py and scripts/video_composer.py) to perform core skill functions. These scripts utilize several third-party libraries to process media files locally.
[EXTERNAL_DOWNLOADS]: The skill instructs users in references/use_cases.md to download and install several standard Python packages (moviepy, pillow, opencv-python, mediapipe, numpy) from public registries. While these are well-known libraries, they represent external dependencies that the skill's main documentation claims to have removed.

video-creation-pro