The Agent Skills Directory

[DYNAMIC_EXECUTION]: The entry-point script scripts/run.py uses os.execv() to replace the current process with a Python interpreter from a dynamically determined virtual environment path (.runners-venv/bin/python3). This mechanism is used for environment management but involves process replacement with an executable at a computed path.
[COMMAND_EXECUTION]: The skill makes extensive use of system commands and subprocesses to run video processing tools (ffmpeg), 3D rendering software (Blender), and various internal Python modules. It invokes these tools via shell execution to perform tasks like video concatenation, audio mixing, and clay blocking renders.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from topics, research briefs, and pre-written scripts, which are then interpolated into complex prompt chains for multiple AI models.
Ingestion points: Inputs are accepted via --research <path> and --topic "<text>" (documented in SKILL.md).
Boundary markers: The instructions do not specify the use of clear delimiters or instructions for the models to ignore embedded directives in the ingested content.
Capability inventory: The skill has high-privilege tool access, including Bash, Write, Edit, Read, and the ability to execute complex media processing pipelines.
Sanitization: While some sanitization is mentioned for ffmpeg filters (escaping colons and commas), there is no documented sanitization or validation for the natural language content used in prompt interpolation.
[DATA_EXFILTRATION]: For specific provider routes (such as OpenRouter), the skill is designed to upload media assets (voice references and images) to litterbox (a temporary file hosting service) to generate public URLs required by the provider APIs. While functional for the task, this involves sending local media components to an external third-party domain.

reel-builder