The Agent Skills Directory

External Content Retrieval: The skill fetches information from various external sources to answer user questions. These sources include Microsoft documentation and repositories, as well as industry expert blogs. While this design ensures users receive authoritative guidance, it involves network operations to domains outside the primary trusted list.
Potential for Indirect Prompt Injection: The skill ingests data from external URLs which could potentially contain instructions intended to influence the agent's behavior. The skill incorporates mitigation strategies by instructing the agent to extract only relevant sections and to cite sources, which helps maintain context and grounding.
Mandatory Evidence Chain for Indirect Prompt Injection: 1. Ingestion points: External content is fetched dynamically based on the topic-to-URL routing table in SKILL.md. 2. Boundary markers: The instructions explicitly direct the agent to 'extract only the section relevant to the question' rather than summarizing whole pages. 3. Capability inventory: The skill utilizes network fetch capabilities; no persistent file-write or arbitrary command execution capabilities were identified. 4. Sanitization: The skill relies on specific extraction and synthesis instructions rather than explicit sanitization or filtering of the fetched content.

eval-faq