The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes content from untrusted local HTML files or remote URLs and uses that content as input for AI-driven code generation.
Ingestion points: The target file path or URL provided to the upgrade, iterate, and screenshot subcommands in runtime/bin/fal-site.mjs.
Boundary markers: The skill uses BEGIN_HTML and END_HTML markers in prompts (found in runtime/src/upgrade.mjs and runtime/src/review.mjs) to delimit user-provided code, but these are not robust against adversarial content.
Capability inventory: The skill can execute shell commands via ffmpeg (runtime/src/video.mjs), control a browser via puppeteer (runtime/src/upgrade.mjs), perform network requests to fal.ai services (runtime/src/fal.mjs), and write files to the local filesystem.
Sanitization: There is minimal sanitization of the ingested HTML content before it is passed to the AI models; the skill primarily relies on basic string cleaning and LLM instruction following.
[COMMAND_EXECUTION]: The skill utilizes the Node.js child_process.spawn function to execute the ffmpeg binary to generate redesign comparison videos as seen in runtime/src/video.mjs.
[EXTERNAL_DOWNLOADS]: The skill communicates with the external fal.run API to perform AI-driven design tasks and downloads generated image assets via the fetch API in runtime/src/fal.mjs. These are vendor-owned resources used for the skill's primary functionality.

fal-redesign