The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses a bash-based probe to discover the installation path of its sibling dependency, ofox-video-core, by checking multiple relative and platform-specific directories (~/.agents, ~/.claude, etc.). Once found, it executes the shell script ofox-video.sh from that path for core operations.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted text data from user-supplied screenshots to populate its video generation prompts, which are then sent to a downstream model.
Ingestion points: Two user-provided screenshot images (before and after states) defined in SKILL.md.
Boundary markers: The skill provides a structured prompt template using explicit section headers (SCREEN, CHANGE, VIEW, ENDING, AVOID) to define model constraints and isolate variable data.
Capability inventory: The skill invokes a local shell script that performs network requests via curl and processes JSON via jq to interact with the ofox.ai API.
Sanitization: The instructions recommend quoting extracted values verbatim; while section headers are used, there is no explicit instruction to sanitize or escape text content extracted from the images before interpolation.

product-demo