Playwright Browser Automation
Warn
Audited by Gen Agent Trust Hub on Aug 29, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
- [DYNAMIC_EXECUTION]: The
run.jsscript implements a custom execution wrapper that accepts arbitrary strings from command-line arguments, files, or standard input. It writes these strings to a temporary JavaScript file in the skill directory and executes them using the Node.jsrequire()function. This enables the execution of dynamically generated automation logic at runtime via computed paths. - Evidence:
run.jslines 180-184:fs.writeFileSync(tempFile, code, 'utf8'); require(tempFile); - [REMOTE_CODE_EXECUTION]: The skill automates the installation of dependencies and browser binaries from the official npm registry and the Playwright project's distribution servers.
- Evidence:
run.jslines 42-43:execSync('npm install', ...); execSync('npx playwright install chromium', ...); - Evidence:
package.jsonscripts:setupandinstall-all-browsers. - [INDIRECT_PROMPT_INJECTION]: The skill processes content from arbitrary external websites provided via user input or auto-detection. This creates an attack surface where malicious content on a webpage could attempt to influence the agent's behavior during the automation process.
- Ingestion points:
page.goto(TARGET_URL)in generated scripts and examples inSKILL.md, as well as link extraction and table data extraction helpers. - Boundary markers: The skill does not implement explicit delimiters or instructions to ignore embedded commands within the web content being processed.
- Capability inventory: The skill has the ability to write files to
/tmp, execute shell commands via therun.jswrapper, and perform network requests via the Playwright browser. - Sanitization: There is no evidence of sanitization or filtering of content scraped from external pages before it is processed by the agent.
- [METADATA_POISONING]: The skill contains inconsistent author information across its metadata files, which could be misleading regarding its origin.
- Evidence:
SKILL.mdfrontmatter listsauthor: Claude Assistant. - Evidence:
package.jsonlists"author": "lackeyjb".
Audit Metadata