The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local utility scripts using Node.js, such as node $A/bin/archify.mjs. These scripts are located within the skill's local directory path defined as ~/.claude/skills/archify.
[COMMAND_EXECUTION]: The skill automates a local web browser (Chrome or Chromium) to perform visual checks. It includes documentation for the ARCHIFY_CHROME_NO_SANDBOX environment variable to disable the Chrome sandbox, which is a common requirement for headless execution in CI/CD containers; the documentation correctly warns against using this flag on personal machines.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and render user-provided HTML and JSON files (out/workflow.html, candidate.json). Processing external artifacts in a browser presents a surface for indirect injection if those files contain malicious scripts or instructions intended to influence the agent's observation of the results.
Ingestion points: Files located at out/workflow.html and candidate.json are read into the tool.
Boundary markers: None identified in the provided instructions for the data files.
Capability inventory: The skill has the capability to execute shell commands via node and manage browser processes.
Sanitization: The tool performs structural and layout validation (e.g., checking scrollWidth and scrollHeight), which provides a layer of validation for the artifacts being processed.

dym-archify-visual-check-and-preview