The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves analyzing untrusted data from external websites (HTML, JavaScript bundles, CSS), which could contain malicious instructions.
Ingestion points: External data is ingested via tools such as analyze_bundle, analyze_stack, analyze_scroll_map, and record_site_motion.
Boundary markers: The skill does not define specific delimiters or instructions to prevent the agent from obeying commands potentially embedded in the analyzed web content.
Capability inventory: The agent has the ability to write to the local file system (e.g., saving video and script chunks) and execute shell commands (ffmpeg, node).
Sanitization: Ingested content is processed for technical features (like animation triggers) but is not explicitly sanitized for prompt injection attacks.
[COMMAND_EXECUTION]: The skill provides instructions for the agent to execute shell commands and local scripts to process video data.
Evidence: Provides ffmpeg command strings for generating image tiles from recordings (ffmpeg -i reference.webm -vf ...).
Evidence: Instructs the agent to run Node.js scripts (node scripts/record-scrollthrough.mjs) from a local repository checkout.
[EXTERNAL_DOWNLOADS]: The skill references an external repository for necessary scripts and dependencies.
Evidence: Directs the agent to the author's GitHub repository https://github.com/INSANE0777/Awwwards-mcp for recording scripts.
[DYNAMIC_EXECUTION]: The skill generates executable code stubs based on parameters extracted from external sites.
Evidence: The build_gsap_stub tool creates a runnable JavaScript stub using durations, easings, and triggers discovered during the site analysis phase.

awwwards-motion-study