The Agent Skills Directory

[PROMPT_INJECTION]: The skill defines a sub-agent architecture in faceless-explainer/sub-agents/frame-worker.md that interprets and builds scenes based on instructions in STORYBOARD.md. This represents an indirect prompt injection surface where malicious instructions within the storyboard could influence the sub-agent's behavior.
Ingestion points: STORYBOARD.md processed by the frame worker.
Boundary markers: Absent; there are no explicit delimiters or instructions for the agent to ignore potentially adversarial content in the storyboard.
Capability inventory: The skill environment allows file system writes and command execution via the hyperframes CLI.
Sanitization: Narrative content processed by the sub-agent is not sanitized before interpretation.
[COMMAND_EXECUTION]: The render-theme.sh script executes a project-local shell script named _postfx.sh to apply plate reaction effects. This allows for the execution of arbitrary commands if an untrusted project directory is processed.
[COMMAND_EXECUTION]: Multiple scripts in the pipeline (e.g., matte.cjs, audio-envelope.cjs, render-and-composite.sh) execute system binaries like ffmpeg and ffprobe to process media files within the project directory.
[EXTERNAL_DOWNLOADS]: The skill fetches and installs dependencies using npx skills add from trusted sources like GitHub organizations heygen-com and vercel-labs. It also utilizes well-known CDNs including jsdelivr.net, unpkg.com, and cdnjs.com for library loading, which are considered safe.
[SAFE]: The use of base64 data-URIs for inlining fonts in build-fonts-css.cjs is a legitimate functional requirement for ensuring deterministic rendering without external resource dependencies.
[SAFE]: Puppeteer is launched with --disable-web-security in scripts like check-overflow.cjs to facilitate the rendering of local file:// compositions, which is a common and necessary configuration for this type of framework.

hyperframes