The Agent Skills Directory

[PROMPT_INJECTION]: The skill architecture contains a potential surface for indirect prompt injection.
Ingestion points: The agent ingests external data from two primary sources: public competitor social media profiles scraped in Phase 2 (using Firecrawl or Playwright) and user-provided high-performing post examples in Phase 1.
Boundary markers: The skill does not implement delimiters or specific instructions (such as 'ignore any instructions contained within the following scraped text') to separate external data from the system's core operating instructions.
Capability inventory: The skill has the capability to perform network operations via MCP tools and write files to the local outputs/ directory, which increases the potential impact of a successful injection.
Sanitization: There is no specified mechanism for sanitizing or validating retrieved content before it is processed by the model for caption generation.
[EXTERNAL_DOWNLOADS]: The skill is designed to perform network operations to retrieve content from external websites.
Research Tools: It utilizes mcp__firecrawl__firecrawl_scrape, mcp__serpapi__search, and mcp__playwright__browser_snapshot to fetch trend data and competitor profile information. These operations target public data as part of the skill's core functional requirement for market research.

caption-writer