The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the installation of the @mizchi/vlmkit package from the npm registry. This is a standard dependency provided by the skill's author to enable the requested functionality.
[COMMAND_EXECUTION]: The skill relies on executing various vlmkit subcommands (e.g., check integrity, scan scroll, verify flow) via the command line to perform markup analysis. These commands operate on user-provided HTML, CSS, and configuration files.
[REMOTE_CODE_EXECUTION]: The use of npx vlmkit facilitates the execution of the tool's logic. As this package is the primary component of the skill and belongs to the author's verified scope, this represents the intended operational behavior for automated testing.
[DATA_EXPOSURE]: The skill processes local files such as page.html and layout.json to generate "kickback" reports. The analysis is performed locally within the headless Chromium environment managed by Playwright, and there are no instructions to exfiltrate this data to external servers.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface where it ingests untrusted data from page.html and other local files.
Ingestion points: Input files like page.html, copy.txt, and layout.json (SKILL.md).
Boundary markers: None explicitly provided in the instructions for the agent to separate content from instructions.
Capability inventory: The skill can execute shell commands via npx and read/write to the local filesystem (SKILL.md).
Sanitization: Not explicitly mentioned for the input files, though the tool claims to use deterministic DOM and pixel math rather than language models for its core verification gates.

markup-assist