The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to interact with arbitrary external websites, which presents a potential surface for untrusted data ingestion.
Ingestion points: External web content is loaded into the agent session via commands like pwcli open, and data is extracted into the context using pwcli snapshot and pwcli eval within SKILL.md and references/cli.md.
Boundary markers: No specific boundary markers or instructions to disregard text-based directives within web pages are specified in the prompts.
Capability inventory: The agent possesses command-line execution capability via scripts/playwright_cli.sh to automate browser operations.
Sanitization: Web page text and DOM snapshots are processed directly as retrieved from the target URLs without sanitization filters.
Terminal Command Execution: The core functionality relies on running local shell commands via the bundled scripts/playwright_cli.sh script to invoke browser operations from the terminal environment.
External Package Retrieval: The wrapper script utilizes npx to dynamically fetch and run @playwright/cli from the public npm registry, which is standard behavior for Node.js-based terminal tools to ensure the required binaries are available.

playwright