The Agent Skills Directory

[DYNAMIC_EXECUTION]: The run.js script dynamically generates a temporary JavaScript file and executes it using the require() function. This mechanism allows the execution of arbitrary code within the skill's environment.\n
Evidence: run.js constructs a temporary file path .temp-execution-${Date.now()}.js, writes code to it using fs.writeFileSync, and then executes it with require(tempFile).\n- [UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill executes shell commands to install dependencies and manage the browser environment, and it provides instructions for running ad-hoc scripts from temporary directories.\n
Evidence: run.js uses child_process.execSync to run bun install and bunx playwright install chromium. SKILL.md instructs the user to write and execute scripts from /tmp/playwright-test-*.js.\n- [DATA_EXPOSURE_AND_EXFILTRATION]: The skill provides helpers for handling user credentials and performing network operations, which could be exploited for data exposure or exfiltration.\n
Evidence: lib/helpers.js contains an authenticate function that processes a credentials object containing sensitive information like passwords. Playwright's automation features can be used to send extracted data to external URLs.\n- [INDIRECT_PROMPT_INJECTION]: The skill's primary function is browser automation, which involves processing data from external websites that may contain malicious instructions.\n
Ingestion points: External web content is ingested via page.goto() in lib/helpers.js and run.js.\n
Boundary markers: None identified; the agent is not instructed to ignore embedded instructions in page content.\n
Capability inventory: Shell command execution (execSync), file system access (fs), and network access via Playwright.\n
Sanitization: No sanitization of ingested content is performed before processing.

playwright-skill