The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data which creates a potential surface for indirect prompt injection.
Ingestion points: The skill reads user-supplied HTML files, exported pages, and live website references (e.g., Framer marketplace URLs) to extract interactions and animations.
Boundary markers: While instructions mention including reference boundaries in the generated output, there are no explicit boundary markers or safety warnings used when the agent initially ingests and parses the source HTML or scripts.
Capability inventory: The agent has the capability to read local files, capture browser-based screenshots and video, execute shell commands (git, ffprobe), and commit changes to a repository.
Sanitization: There is no evidence of sanitization or filtering of the source HTML/JS content before it is processed by the agent.- [COMMAND_EXECUTION]: The skill workflow involves executing shell commands to manage files and verify assets.
Evidence: Uses git status, git diff, git add, and git commit to manage the article worktree. It also uses ffprobe to verify the integrity and content of recorded MP4 video files.

html-to-interaction-prompts