marketing-launch
Warn
Audited by Gen Agent Trust Hub on Mar 13, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill attempts to globally install the
agent-browserpackage andplaywrightchromium binaries if they are missing from the system. - [REMOTE_CODE_EXECUTION]: Shell commands are used to install external dependencies from npm and npx registries, which involves fetching and executing third-party code at runtime.
- [COMMAND_EXECUTION]: The skill executes various shell commands to manage browser sessions and perform research on external websites like Product Hunt, LinkedIn, and Reddit.
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted web content (e.g., social media posts, community forums, and competitor blogs) and uses this data to inform and implement changes in the local repository codebase. * Ingestion points: Content scraped from Product Hunt, Reddit, LinkedIn, and competitor URLs via
agent-browserinSKILL.md. * Boundary markers: There are no instructions to the agent to ignore or delimit potentially malicious instructions embedded in the scraped web data. * Capability inventory: The skill has the authority to read, recommend, and implement changes directly into the user's local codebase in 'Implementation Mode'. * Sanitization: No sanitization, validation, or filtering mechanisms are applied to the external web content before it is used for code generation.
Audit Metadata