The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill uses Playwright to ingest and analyze arbitrary content from external website URLs provided at runtime. This creates a vulnerability where a malicious website could contain hidden instructions designed to manipulate the agent's behavior or exfiltrate processed data.
Ingestion points: Comprehensive website scraping via analyze-web.js using Playwright.
Boundary markers: The skill documentation does not mention any delimiters or instructions to ignore embedded commands within the scraped content.
Capability inventory: The agent is authorized to execute shell scripts, write to the filesystem, and perform network requests to external APIs.
Sanitization: There is no mention of sanitizing or validating the extracted website data before it is used to drive the creative and decision-making logic.
[COMMAND_EXECUTION]: The skill is configured to run a variety of local shell, Python, and JavaScript scripts (generate-slides.sh, publish-carousel.sh, analyze-web.js, etc.) to perform its core functions. The 'Zero Confirmation' rule explicitly instructs the agent to execute these commands autonomously without user approval.
[PERSISTENCE]: The skill implements an autonomous persistence mechanism via its 'Self-Schedule' feature. It reads performance data from a local file and is instructed to schedule its own next execution, allowing it to maintain an active, recurring presence on the system without user intervention.
[EXTERNAL_DOWNLOADS]: The skill documentation indicates a requirement to download and install external binaries (Chromium) via the Playwright command-line interface to support its web analysis features.

marketing-carousel-growth-engine