The Agent Skills Directory

[PROMPT_INJECTION]: The skill has an indirect prompt injection surface because it processes untrusted data from live product DOMs, console outputs, and source code repositories (stations 3, 7, 8, and 9). The instructions lack formal boundary markers or sanitization procedures for this external content. Although the 'Evidence before judgment' rule and [verified] tags help classify data, the agent's ability to execute shell commands and interact with browsers increases the potential risk if malicious instructions are encountered in audited materials.
[COMMAND_EXECUTION]: The skill directs the agent to run command-line tools for technical auditing, including 'npx lighthouse', 'npm audit', and 'npm test'. These are standard procedures for product inspection and involve well-known, trusted utilities.
[EXTERNAL_DOWNLOADS]: The instructions encourage the use of trusted third-party tools and libraries for performance and accessibility testing, such as Lighthouse, axe-core, and pa11y, which are widely accepted in the software development community.

product-inspection