The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion and analysis of external, untrusted data from various sources such as patent databases, research papers, and web search results, creating a potential attack surface for indirect prompt injection.
Ingestion points: External content fetched via tools like Exa, Tavily, Brave, and generic web search, as specified in references/source-routing.md.
Boundary markers: The instructions do not explicitly require the agent to utilize specific boundary markers or 'ignore instructions' prompts when processing retrieved text, which could lead to the agent following malicious instructions hidden in technical documents or web pages.
Capability inventory: The agent has the ability to write to the local filesystem to generate reports and logs (SKILL.md, references/workflow.md) and call network search tools.
Sanitization: There is no evidence of content sanitization or validation routines for the data retrieved from external sources before it is processed by the agent.
[DYNAMIC_EXECUTION]: The documentation in references/deliverables.md suggests using external rendering tools such as pandoc, python-docx, and reportlab. While these are optional suggestions for the host environment rather than direct skill commands, they represent a path for the dynamic invocation of external binaries and libraries on content generated by the agent.

competitive-landscape