The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using npx, specifically npx prima-cli and npx playwright-cli. These tools are used to control browser sessions and perform automation tasks such as form filling and behavior checking.
[DYNAMIC_EXECUTION]: The pw command accepts a string containing executable code (e.g., ({ page }) => page.click(...)) which is executed within the Playwright browser context. This allows for dynamic code execution based on instructions provided in the prompt.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external websites through commands like check, do, and ask. This presents a surface where malicious instructions on a website could influence the agent's behavior.
Ingestion points: Browser viewport, HTML structure, and screenshots from external web pages processed by the AI layer.
Boundary markers: No explicit boundary markers or sanitization instructions are provided to separate page content from the agent's operational instructions.
Capability inventory: Capability to write files (reports and session artifacts), execute shell commands via npx, and manipulate browser sessions.
Sanitization: There is no mention of filtering or sanitizing the content retrieved from web pages before it is processed by the AI models.
[EXTERNAL_DOWNLOADS]: The skill uses npx to execute tools and mentions npx playwright install chromium, which involves downloading browser binaries from Playwright's infrastructure. These are recognized tools from the vendor's ecosystem.

prima