The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a shell command template npx lighthouse {URL}. If the {URL} parameter is populated with unsanitized user input, an attacker could use shell metacharacters such as ; or && to execute unauthorized commands on the host environment.
[EXTERNAL_DOWNLOADS]: Fetches and runs Google's Lighthouse utility via npx at runtime. The skill also utilizes the @axe-core/playwright and playwright libraries for accessibility testing and browser automation. These are well-known tools from trusted organizations (Google, Deque Systems, and Microsoft).
[PROMPT_INJECTION]: The skill processes untrusted data from external websites, creating an indirect prompt injection surface where malicious site content could influence the agent's summary reports or subsequent decisions. Evidence Chain: 1. Ingestion points: Fetches content from external URLs via Lighthouse and Playwright. 2. Boundary markers: No markers are defined to separate external content from the agent's instructions. 3. Capability inventory: Subprocess execution (npx), file system writes (qa-reports/), and network access. 4. Sanitization: No explicit sanitization or validation of the URL or website content is implemented in the instruction set.

web-audit