The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from source Markdown files and target websites to drive browser actions and image generation prompts.
Ingestion points: Markdown files in the cases/ directory and content/DOM of target websites visited via Playwright.
Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are implemented for the processed text.
Capability inventory: The skill can perform browser interactions (click, type, navigate) via Playwright MCP and network requests to OpenRouter via a bundled Python script.
Sanitization: No explicit sanitization or validation of the input strings is performed before they are used as prompts or interaction targets.
[EXTERNAL_DOWNLOADS]: The skill downloads the Playwright MCP server and Chromium browser binaries from official package registries.
Evidence: Setup instructions in references/mcp-setup.md reference npx @playwright/mcp@latest and npx playwright install chromium.
[COMMAND_EXECUTION]: The skill executes shell commands to install dependencies and run a local Python script for image generation.
Evidence: Usage of python scripts/generate_image.py and various npx commands. The practice of interpolating user-controlled descriptions into shell command strings without clear escaping creates a potential command injection surface.

doc-snapshot-agent