The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes several external binaries including Chromium-family browsers (Chrome, Canary, Chromium, Edge), ImageMagick (magick), and Node.js to manage the rendering, measurement, and comparison of visual deliverables.
[DYNAMIC_EXECUTION]: Shell scripts such as matrix.sh and render.sh dynamically generate and inject JavaScript probes into local HTML files using inline Python code. Furthermore, converge-driver.mjs uses the Chrome DevTools Protocol (CDP) to evaluate JavaScript expressions within the browser context to detect settled animation states.
[DATA_EXFILTRATION]: The skill configures the browser with the --allow-file-access-from-files flag. While this enables the verification of local artifacts that load modules or assets, it technically broadens the file access capabilities of the headless browser during the render process. Diagnostic data is extracted from the browser via document.title and parsed by shell utilities.
[INDIRECT_PROMPT_INJECTION]: The skill exposes an attack surface where untrusted data (HTML content) could influence agent behavior, though this is a low-risk diagnostic use case.
Ingestion points: Local HTML files and specified URLs processed by render.sh, matrix.sh, and converge.sh (via converge-driver.mjs).
Boundary markers: No specific delimiters or "ignore" instructions are used when interpolating content into the rendering pipeline, though the skill operates on discrete files rather than continuous conversational context.
Capability inventory: Process spawning (browser, magick, node, python3), file system writes (screenshots, reports, manifest files), and local network communication (CDP WebSocket on 127.0.0.1).
Sanitization: The checks.js script employs a safe() function to sanitize text strings before they are exported to document.title, preventing malformed data from affecting the shell-based parsing logic.

visual-verify