The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from untrusted external URLs during the motion study and bundle analysis phases.
Ingestion points: Website content is processed via record_site_motion, analyze_scroll_map, and analyze_bundle (which fetches script chunks).
Boundary markers: No specific boundary markers or "ignore instructions" warnings are defined for the ingested website content.
Capability inventory: The skill can execute local scripts (node scripts/...), run CLI tools (ffmpeg), and generate code stubs (build_gsap_stub).
Sanitization: No explicit sanitization of external website content or bundle data is mentioned before processing.
[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands and local scripts.
Evidence: Usage of ffmpeg for video tiling (ffmpeg -i reference.webm -vf ...).
Evidence: Execution of local Node.js scripts provided with the skill: scripts/record-scrollthrough.mjs and scripts/skill-memory.mjs.
[DYNAMIC_EXECUTION]: The skill generates executable code stubs based on analyzed data.
Evidence: The build_gsap_stub tool generates a runnable GSAP+ScrollTrigger stub using parameters (durations, easings, etc.) extracted from the untrusted external website.
Evidence: The analyze_bundle tool fetches and optionally de-obfuscates external JavaScript chunks using webcrack.
[EXTERNAL_DOWNLOADS]: The skill fetches content and script chunks from external websites for analysis.
Evidence: analyze_bundle fetches site script chunks (up to 40 chunks) from external domains to perform bundle forensics.

awwwards-motion-study