The Agent Skills Directory

[COMMAND_EXECUTION]: Several Python scripts in the scripts/ directory (prepare_evidence.py, materialize_reverse_media.py, digest.py) utilize subprocess.run() to invoke local ffmpeg and ffprobe binaries. These tools are used for their primary intended purpose: extracting video metadata, capturing frames, and generating low-bitrate previews for analysis and review.
[DYNAMIC_EXECUTION]: The scripts/digest.py script generates complex HTML fragments containing embedded JavaScript. This JavaScript is designed to provide an interactive user interface within the AI agent's visualization environment, enabling features such as synchronized video playback, tabbed navigation between analysis sections, and the ability to drag media previews directly into other applications.
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests untrusted data from local video files to generate prompts for other models.
Ingestion points: scripts/prepare_evidence.py extracts frames and metadata from user-provided video files.
Boundary markers: The skill includes a dedicated analysis prompt (references/analysis-prompt.md) that explicitly instructs the analysis model to focus only on observable facts from the video and to output data in a strict JSON schema format.
Capability inventory: The skill has capabilities for file system read/write and local command execution via ffmpeg and ffprobe.
Sanitization: Validation is performed using a strict JSON schema (schema.json) and a dedicated validation script (scripts/validate.js) using the ajv library to ensure structural integrity and check for suspicious placeholder text.
[EXTERNAL_DOWNLOADS]: The skill uses standard, well-known dependencies such as the ajv package for Node.js and the jsonschema library for Python to perform data validation. It also references the use of platform-provided multimodal models like Gemini for semantic video analysis, which is standard behavior for analytical skills.

100x-video-reverse