The Agent Skills Directory

[PROMPT_INJECTION]: The skill documentation includes a mandatory alignment step where the agent must confirm the duration, content, and consistency anchors with the user before proceeding to generation. This workflow acts as a verification layer to prevent the agent from executing unintended or malicious instructions hidden in vague user requests.- [PROMPT_INJECTION]: An indirect prompt injection surface is present because the skill takes user-provided descriptions to build prompts for external video generation models.
Ingestion points: User-provided video descriptions and content details defined in the 'Step 1 — Align scope' and 'Step 2 — Write the prompt' sections of SKILL.md.
Boundary markers: No explicit delimiters or boundary instructions are defined to separate user input from system instructions when formatting the final prompt for the video models.
Capability inventory: The skill utilizes submit_video to trigger external generation jobs and edit_item / manage_timelines to modify the project library and timeline based on the model's output.
Sanitization: The instructions do not specify a process for sanitizing, escaping, or validating the content of the user-provided prompt before interpolation into the tool parameters.- [REMOTE_CODE_EXECUTION]: No evidence of unauthorized remote code execution, piped shell commands, or unverifiable dependency installations was found. The skill uses managed platform tools for all operations.- [DATA_EXFILTRATION]: The skill logic is confined to processing media within the project library and communicating with authorized video generation endpoints. No patterns indicating unauthorized data harvesting or exfiltration were identified.

video-gen