The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted text from the user (e.g., articles or notes) into capture/extracted/visible-text.txt. This content is later processed by the agent to create a storyboard and passed to sub-agents for frame construction.
Ingestion points: capture/extracted/visible-text.txt, BRIEF.md, and user_script.txt (SKILL.md Step 1).
Boundary markers: The workflow references ../hyperframes/references/subagent-dispatch.md and ../hyperframes/references/frame-worker-core.md to guide sub-agent behavior and boundaries.
Capability inventory: The skill uses spawnSync to execute media tools (ffmpeg, ffprobe) and utilizes the npx CLI for project management and rendering tasks.
Sanitization: There is no explicit sanitization of the ingested text in the provided scripts; the skill relies on the agent's instructions and platform-level dispatch guardrails to handle the content safely.
[DYNAMIC_EXECUTION]: The skill dynamically assembles 'frame packets' for sub-agents in scripts/lib/frame-packets-core.mjs. These packets include storyboard blocks, blueprint bodies, and motion recipes which are then used by sub-agents to generate HTML/JS compositions. While this is the primary function of the skill, the runtime assembly of executable instructions represents a dynamic execution vector.
[COMMAND_EXECUTION]: The skill executes local commands using child_process.spawnSync. Specifically, scripts/assemble-index.mjs and scripts/audio.mjs invoke ffmpeg and ffprobe for audio processing and looping. The skill employs the -- argument separator as a security best practice to mitigate potential command-line option injection from untrusted filenames.

faceless-explainer