The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted HTML, CSS, and JS snapshots as input. There is a risk that malicious instructions embedded within these source files could be interpreted by the LLM during the extraction (Wave 0/1) or build (Wave 2/4) phases, potentially altering the intended output or behavior. * Ingestion points: source-pages/*.html, _files/*.css, and _files/*.js are read and parsed. * Boundary markers: While the skill uses structured 'waves' and 'briefs', it lacks explicit delimiters or instructions to ignore potential commands embedded in the snapshot content itself. * Capability inventory: The pipeline involves file system access, network downloads (via curl for assets), and the execution of npm install and next build on generated code. * Sanitization: There is no documented process for sanitizing or escaping the content extracted from the snapshots before it is used to generate documentation or code.
[COMMAND_EXECUTION]: The skill instructions include shell commands (e.g., find, grep, sed, curl) to be executed as part of the pipeline. While these are standard tools, they are used to process and download data based on the content of external snapshots.
[EXTERNAL_DOWNLOADS]: The skill automatically identifies and downloads external assets (fonts, images, icons) referenced in the snapshots' HTML and CSS files using curl. This behavior is central to its purpose but involves interaction with arbitrary external URLs found in the source data.

snapshot-to-nextjs