The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill utilizes web research to gather benchmarks and analogs. It includes specific instructions to use generic queries for proprietary details and to avoid requesting or storing sensitive personal data, which mitigates data exposure risks.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the public web during research phases defined in Step 1 and Step 3. While this creates a vulnerability surface, the skill lacks high-privilege capabilities like code execution or file modification, and it instructs the assistant to summarize and paraphrase rather than directly inject external content.
[PROMPT_INJECTION]: The skill defines its own operating principles and a conversation contract, but these are instructional and do not attempt to bypass or override the agent's core safety guidelines. It explicitly includes an ethics and safety section.

scamper