The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions in SKILL.md (Phase 2, Step 3) direct the agent to verify image URLs by running the shell command curl -sI "<url>" | head -1. Because the <url> is sourced from external web search results (such as Pixabay or Unsplash), it could contain shell metacharacters (e.g., ;, &, |) intended to execute arbitrary code on the host system.
[INDIRECT_PROMPT_INJECTION]: The skill relies on external data sources that could be manipulated by an attacker to influence the agent's behavior or compromise the system environment.
Ingestion points: The agent fetches statistics and image URLs from external web content during Phase 2.
Boundary markers: There are no protective boundary markers or instructions to treat the research data as untrusted content.
Capability inventory: The skill utilizes shell command execution via curl and has access to the local file system for template loading.
Sanitization: The skill does not provide instructions for sanitizing or validating the image URLs before they are interpolated into a shell command.

blog-write