The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the subprocess.run method to invoke external command-line utilities essential for media production. These include ffmpeg for merging audio and video, ffprobe for identifying file properties, and yt-dlp for retrieving video content. Additionally, fetch_footage.py contains logic to self-update yt-dlp using the -U command if the tool encounters errors related to being outdated.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to obtain resources from the internet. It retrieves video footage from YouTube using yt-dlp, fetches summary text from the Wikipedia REST API via urllib, and allows for capturing screenshots of web pages through Playwright.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted text and metadata from external sources (Wikipedia, YouTube) to help the agent draft video scripts. This creates a potential vulnerability where malicious instructions hidden in the source content could be interpreted by the LLM as commands.
Ingestion points: Wikipedia REST API summary extracts, YouTube search results/titles, and web page content captured via Playwright.
Boundary markers: Absent; there are no specific delimiters defined to separate fetched content from the agent's internal instructions.
Capability inventory: File system access, network requests, and shell command execution for media processing.
Sanitization: Absent; retrieved text is used directly in prompts for script generation without sanitization or filtering for adversarial patterns.
[DYNAMIC_EXECUTION]: The skill's primary operation involves the agent generating a Python rendering script (render.py) based on a provided template and then executing that script to generate video frames. This dynamic generation of executable code is a fundamental part of the construction process.

fireship-style-edit