The Agent Skills Directory

[SAFE]: The skill contains no executable code or scripts. All files provided (SKILL.md, evals.json, source-guides.md) are instructional Markdown or JSON definitions.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for the agent to use its browsing capabilities to research well-known platforms such as Reddit, G2, LinkedIn, and Hacker News. These references are for legitimate research purposes and do not involve the download or execution of untrusted software.
[INDIRECT_PROMPT_INJECTION]: The skill instructions involve analyzing untrusted data from public forums and review sites. • Ingestion points: External community posts and reviews (e.g., Reddit, G2, YouTube comments) are processed as part of Mode 2 research. • Boundary markers: Not explicitly defined in the prompt instructions, though the skill focuses on verbatim quote extraction rather than instruction following. • Capability inventory: The skill has no capabilities to execute commands, modify system files, or perform network operations beyond general browsing. • Sanitization: No explicit sanitization of research data is mentioned, but the risk is low given the lack of exploitable capabilities.

customer-research