The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes subprocess.run within its internal scripts (e.g., audit_ui.py, finalize_visual_review.py) to coordinate the execution of helper tools. The shared implementation in _skill_cli.py employs security best practices by using shell=False and resolving executables via shutil.which to prevent command injection vulnerabilities.
[DYNAMIC_EXECUTION]: The audit_contrast.py script executes a JavaScript probe (contrast_probe.js) within the Playwright browser context via page.evaluate. This dynamic execution is essential for calculating real-time WCAG contrast ratios and evaluating motion-over-text risks directly in the browser.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon data from local project documentation files.
Ingestion points: Reads product.md, design.md, and mockup-brief.md in load_visual_context.py to extract route intent and design specifications.
Boundary markers: No specific delimiters or boundary markers are used when processing the content of these files.
Capability inventory: The agent has the capability to execute subprocesses (via _skill_cli.py) and perform file-write operations (via append_fix_log.py).
Sanitization: There is no evidence of sanitization or filtering of the ingested markdown content before it is used by the agent to make code-fix decisions.
[EXTERNAL_DOWNLOADS]: The skill documentation and scripts (e.g., audit_ui.py) reference the installation of playwright and its browser binaries using standard tools like pip, uv, and the Playwright CLI. No unverified or malicious remote code execution patterns were identified.

visual-auditor