playwright
Fail
Audited by Gen Agent Trust Hub on Mar 18, 2026
Risk Level: HIGHCREDENTIALS_UNSAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill instructs the agent to source credentials from
~/.config/taledraw-test.envand retrieve passwords from the macOS Keychain usingsecurity find-generic-password -a talelens-test -s <service> -w. This creates a direct path for the agent to access and potentially expose sensitive user data. - [COMMAND_EXECUTION]: The skill relies on a bash wrapper script
scripts/playwright_cli.shto execute browser automation tasks. This script handles command-line arguments and executes the Playwright CLI, which can perform high-privilege browser operations and execute JavaScript viaevalandrun-codesubcommands. - [EXTERNAL_DOWNLOADS]: The wrapper script uses
npx --package @playwright/cli playwright-clito download and execute code from the NPM registry at runtime. While Playwright is a well-known tool from Microsoft, fetching and executing remote packages introduces a dependency on external registry availability and integrity. - [PROMPT_INJECTION]: The skill is highly vulnerable to indirect prompt injection because it is designed to ingest and process content from arbitrary websites. Malicious instructions on a target web page could be interpreted by the agent during snapshots or evaluations.
- Ingestion points: Web content processed via
snapshotandevalcommands inscripts/playwright_cli.shand documented inreferences/cli.md. - Boundary markers: The instructions do not define markers to separate untrusted web data from the agent's system instructions.
- Capability inventory: The agent has access to terminal execution, browser control, and file system access (writing to
output/playwright/). - Sanitization: There is no evidence of sanitization or filtering of the DOM content or JavaScript evaluation results before they are returned to the agent's context.
- [DATA_EXFILTRATION]: The combination of accessing Keychain credentials and having the capability to interact with external websites via a browser creates a significant exfiltration surface if the agent is compromised by malicious instructions.
Recommendations
- AI detected serious security threats
Audit Metadata