The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill generates media based on user-supplied prompts which are passed to external CLI tools like fal.mjs. This creates an attack surface where malicious input could influence the behavior of the media generation model or the shell commands if not properly sanitized.
Ingestion points: User-provided prompts are passed as arguments to node $F image, node $F edit, and node $F video in SKILL.md.
Boundary markers: The skill does not explicitly define delimiters or instructions to ignore embedded instructions within the user-provided prompts.
Capability inventory: The skill executes shell commands using node, bash, python3, and ffmpeg to process media and interact with the fal.ai API.
Sanitization: No explicit sanitization, escaping, or validation logic is documented for the prompts interpolated into shell commands.
[COMMAND_EXECUTION]: The skill relies on executing local scripts (fal.mjs, prep-footage.sh, shots.py) and system tools like ffmpeg via the shell. While these scripts are internal to the skill's environment (referenced via ${CLAUDE_PLUGIN_ROOT}), the direct interpolation of user-supplied prompts into shell command arguments is a common vulnerability surface that requires robust handling by the agent platform.

hyperframe-pro-footage