The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill interacts with external website content via page.goto() and page.evaluate(), creating a surface where malicious web content could influence agent behavior.
Ingestion points: Browser navigation and DOM evaluation in SKILL.md, references/chatgpt-dalle.md, and references/chinese-login-modals.md ingest untrusted data from the open web.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands within the scraped content were identified.
Capability inventory: The skill utilizes subprocess execution (ssh, scp, pip), file system writes (saving cookies and images to /tmp/), and network operations via Playwright.
Sanitization: Web content is processed directly for title extraction and button clicking without specific sanitization filters.
[CREDENTIALS_UNSAFE]: The skill provides templates and instructions for handling sensitive authentication data.
Evidence: Code snippets in SKILL.md and references/vultr-login.md guide the user to input usernames and passwords into automation scripts.
Vendor Resources: The account 1351712821@qq.com and IP 45.76.185.1 are provided as configuration references within the author's context.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of external dependencies for its core functionality.
Evidence: Instructions in SKILL.md specify installing the playwright Python package and the chromium browser via npx playwright install.
[COMMAND_EXECUTION]: The skill utilizes shell commands for environment setup and remote task execution.
Evidence: Usage of pip3 install, npx playwright, ssh, scp, and pkill to manage the browser environment and execute scripts on remote servers.
[DYNAMIC_EXECUTION]: The skill frequently uses page.evaluate() to execute JavaScript within the browser context to bypass UI overlays and interact with elements.
Evidence: JavaScript injection patterns are used for clicking buttons and extracting image data as base64 strings.

browser-automation