The Agent Skills Directory

[DATA_EXFILTRATION]: The plan.py script accesses the sensitive credential file at ~/.config/cartesia/api_key to retrieve API credentials, which are subsequently transmitted to the external endpoint api.cartesia.ai within the Authorization header.
[COMMAND_EXECUTION]: The skill utilizes subprocess.run to execute several system-level tools, including ffprobe for media analysis and various headless browser binaries (e.g., Chromium, Google Chrome) for PDF generation. It also instructs the agent to run the pdf2md and cartesia CLI utilities for document conversion and voice synthesis.
[DYNAMIC_EXECUTION]: The skill implements a custom markdown-to-HTML parser in plan.py and renders the output using a headless browser with the --allow-file-access-from-files flag enabled. This configuration permits the browser to access local filesystem resources, which could be exploited if malicious content is processed through the video plan template.
[INDIRECT_PROMPT_INJECTION]: The skill presents an indirect prompt injection surface when processing external data:
Ingestion points: The skill reads untrusted client data from the brief/ folder (including converted PDFs and chat messages) and reference video metadata.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to ignore potentially malicious embedded instructions within the client briefs.
Capability inventory: The skill possesses the capability to write files, execute shell commands via subprocess, and perform network operations.
Sanitization: No sanitization or validation of the ingested text is performed before it is incorporated into the agent's context for plan generation.
[EXTERNAL_DOWNLOADS]: The skill fetches generated voiceover audio data from https://api.cartesia.ai (a well-known voice synthesis service) based on the contents of the user-provided script files.

video-plan