The Agent Skills Directory

[COMMAND_EXECUTION]: The script assemble-index.mjs executes ffmpeg and ffprobe using child_process.spawnSync to perform legitimate media processing tasks, such as extending audio loops to match video length.
[COMMAND_EXECUTION]: The workflow orchestrates several local Node.js scripts and npx commands from the HyperFrames toolchain to manage project initialization, audio syncing, and final rendering.
[PROMPT_INJECTION]: The skill ingests arbitrary user-provided text into visible-text.txt, creating an indirect prompt injection surface. This text is later processed by sub-agents to generate frame content. The skill mitigates risk by using isolated packets for each frame worker, limiting the scope of any potential injection to the HTML output of individual scenes.
[DYNAMIC_CONTEXT_INJECTION]: The SKILL.md file utilizes dynamic injection placeholders (!command) to run status checks and self-updates via the hyperframes CLI. These operations are restricted to static, vendor-provided commands and do not process untrusted external data.

faceless-explainer