carousel-growth-engine
Warn
Audited by Gen Agent Trust Hub on Sep 10, 2026
Risk Level: MEDIUMPROMPT_INJECTIONINDIRECT_PROMPT_INJECTIONTIME_DELAYED_CONDITIONALDYNAMIC_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill contains instructions that significantly increase the agent's autonomy and reduce oversight. It explicitly commands the agent to "never ask for permission between steps", "Zero Confirmation", and "Notify Only at End". These instructions encourage the agent to bypass standard user confirmation checkpoints for high-risk actions such as network publishing and file system modifications.
- [INDIRECT_PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection by processing arbitrary external website content. * Ingestion points: The
analyze-web.jsscript uses Playwright to scrape content from any website URL provided or discovered during the research phase. * Boundary markers: The instructions do not specify any boundary markers or instructions to ignore embedded commands within the ingested web content. * Capability inventory: The skill possesses capabilities for network communication with external APIs (api.upload-post.com), file writing (learnings.json), and autonomous task scheduling. * Sanitization: There are no mentioned mechanisms for sanitizing, filtering, or escaping the data extracted from websites before it is interpolated into slide prompts or platform captions. - [TIME_DELAYED_CONDITIONAL]: The skill implements an autonomous scheduling mechanism that gates future executions based on external data. It instructs the agent to "Self-Schedule" and "set next cron execution at the optimal hour" based on performance insights stored in the local
learnings.jsonfile. This allows the skill to maintain persistence and trigger actions at future times without direct user interaction. - [DYNAMIC_EXECUTION]: The skill utilizes Playwright via
analyze-web.jsto perform deep analysis of websites, which involves rendering and executing JavaScript from untrusted external sources. While Playwright is a standard tool for this task, its use with arbitrary external input combined with the agent's high autonomy poses a risk if the execution environment is redirected or manipulated by the processed content.
Audit Metadata