The Agent Skills Directory

[COMMAND_EXECUTION]: The skill implements a 'Zero Confirmation' policy, executing a sequence of local scripts (generate-slides.sh, analyze-web.js, publish-carousel.sh, check-analytics.sh) without seeking user approval for individual steps. It also features an autonomous persistence mechanism that schedules future executions by modifying system cron jobs based on performance data.\n- [EXTERNAL_DOWNLOADS]: The skill uses Playwright to navigate and scrape arbitrary external websites and downloads browser binaries via playwright install chromium. It also communicates with external APIs at generativelanguage.googleapis.com (Gemini) and api.upload-post.com (Upload-Post).\n- [PROMPT_INJECTION]: The instructions explicitly direct the agent to suppress user confirmation and oversight ('Zero Confirmation', 'never ask for permission'), overriding standard safety behaviors. Additionally, the skill has a significant indirect prompt injection surface due to its autonomous ingestion of untrusted data:\n
Ingestion points: Full scraping of target URLs and internal pages (pricing, testimonials, etc.) via analyze-web.js.\n
Boundary markers: Absent; no delimiters or warnings to ignore instructions within the scraped content are specified.\n
Capability inventory: Execution of shell, JavaScript, and Python scripts; network access to multiple APIs; and direct publishing to social media platforms.\n
Sanitization: Absent; extracted content from websites is directly interpolated into prompts for slide generation without validation or escaping.\n- [DATA_EXFILTRATION]: The automated workflow is designed to move content from arbitrary external sources to a third-party API. This architecture creates a risk of data exfiltration if the agent is manipulated into scraping sensitive local files or environment variables and including them in the data sent to the Upload-Post API.

agency-carousel-growth-engine