The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. It scrapes content from external, untrusted URLs provided by the user (using Firecrawl). If a targeted website contains malicious instructions hidden in its HTML or text, the agent might inadvertently follow those instructions during the synthesis phase.
Ingestion points: External URLs scraped via firecrawl_map, firecrawl_scrape, and firecrawl_search in SKILL.md.
Boundary markers: None identified; the skill processes raw markdown from scrapes.
Capability inventory: File system writes to the competitor-profiles/ directory and network requests via MCP tools.
Sanitization: No explicit sanitization or filtering of scraped content is described before it is processed by the agent.
[DATA_EXPOSURE]: The skill attempts to read local project files such as .agents/product-marketing.md, .claude/product-marketing.md, or product-marketing-context.md to gather context about the user's own product. While this is a documented feature for providing context, users should be aware that the agent accesses these local files.
[COMMAND_EXECUTION]: The skill utilizes several MCP tools (firecrawl_*, dataforseo_*) to perform its tasks. These are executed within the agent's tool-use framework and do not involve direct shell command execution of untrusted input.

competitor-profiling