The Agent Skills Directory

[COMMAND_EXECUTION]: The skill launches the Chromium browser with security-reducing flags, specifically --disable-web-security and --allow-file-access-from-files. These flags disable the Same-Origin Policy and permit access to local files, which significantly increases the risk of data theft if the agent visits a malicious website.\n- [REMOTE_CODE_EXECUTION]: During the first run, the skill automatically executes npm install playwright and npx playwright install chromium to a local cache directory (~/.cache/axure-extractor). While these are standard for Playwright-based tools, automated package installation and binary downloads from remote sources represent a potential supply chain risk.\n- [COMMAND_EXECUTION]: The skill uses new Function() within Playwright's page.evaluate() context to dynamically wrap and execute JavaScript code read from local files (inject/*.js). This dynamic execution pattern is used across multiple modules (links.mjs, markdown.mjs, theme.mjs) to perform extraction logic in the browser context.\n- [PROMPT_INJECTION]: The skill is vulnerable to Indirect Prompt Injection because it ingests and converts untrusted data from external websites into Markdown for the agent to process.\n
Ingestion points: Web page content is extracted in scripts/inject/extract-markdown.js and scripts/inject/extract-links.js.\n
Boundary markers: There are no explicit boundary markers or instructions to the agent to ignore instructions embedded within the extracted Markdown.\n
Capability inventory: The skill can execute shell commands (execSync in browser.mjs), write to the local file system, and perform network requests via the browser.\n
Sanitization: The skill performs direct DOM-to-Markdown conversion without filtering or sanitizing the content for malicious LLM instructions.\n- [EXTERNAL_DOWNLOADS]: The skill fetches the Playwright library and Chromium browser binaries from official registries and established mirrors.

extract-page-data