The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references an installation script located at https://raw.githubusercontent.com/voidmatcha/ui-clone-skills/main/install.sh. Per the author's instructions, this command is surfaced to the user for manual execution if dependencies are missing, rather than being executed automatically by the skill.
[COMMAND_EXECUTION]: Multiple scripts (e.g., ae-compare.sh, section-compare.sh, animation-spec-compare) use subprocess.run or direct shell execution to invoke system utilities such as magick, identify, ffmpeg, dssim, and agent-browser. These operations are fundamental to the skill's purpose of visual analysis.
[DYNAMIC_EXECUTION]: The suite dynamically generates and executes JavaScript snippets within browser sessions via agent-browser --session eval. This is used to extract live metrics like computed styles, DOM nesting, and animation frame progress. The logic is encapsulated in library files like lib/enumerate-sections.js and lib/extract-dom.js.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it ingests and processes untrusted data from external URLs during the verification process.
Ingestion points: ref-url and impl-url parameters in scripts such as live-parity-sweep.sh and section-compare.sh.
Boundary markers: Findings are typically output to structured JSON or markdown files (e.g., sections/result.json), which helps isolate external content from the agent's primary instruction stream.
Capability inventory: The skill can execute shell commands via subprocess and perform network operations through the browser automation tool.
Sanitization: The skill performs normalization on text and CSS properties to facilitate comparison, though it does not explicitly filter for malicious payloads within the analyzed sites.

visual-debug