The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses a potential attack surface through the ingestion of untrusted data when using the hallmark study <URL> or hallmark audit commands.
Ingestion points: The skill reads local project files and fetches remote web pages (HTML/CSS) for design analysis.
Boundary markers: The skill contains explicit instructions in references/study.md and references/verbs/redesign.md to treat fetched content as untrusted inert data and to ignore any instructions found in remote scripts, comments, or metadata.
Capability inventory: The skill has the ability to execute local Python and Node.js scripts provided in the skill package, perform network requests via WebFetch, and write or modify design-related files (tokens.css, design.md).
Sanitization: The instructions specifically mandate that the agent only extract "design facts" and ignore attempts to reveal secrets or override instructions.
[EXTERNAL_DOWNLOADS]: The skill uses WebFetch to download remote content for visual analysis. It also references numerous external asset providers, including Google Fonts, Fontshare, and Unsplash, which are well-known and standard for frontend design tasks.
[COMMAND_EXECUTION]: The skill instructs the agent to run local utility scripts packaged within the skill directory, such as fdk/tools/build-design-showcase.py and fdk/tools/button-harmony.mjs. These are used to automate design auditing and component generation tasks.

hallmark