The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a Python script scripts/check-output.py which is a benign utility designed to validate the existence of specific markdown sections in generated output files. It does not perform network operations or access sensitive system data.
[DATA_EXFILTRATION]: Research prompts such as P01 (Account Snapshot) and P02 (Work Email Find) are designed to process public business information. These prompts include strict rules requiring source URLs for all claims and explicitly forbid pattern-guessing, with no indicators of unauthorized data transfer.
[PROMPT_INJECTION]: Templates that process external data, such as email bodies and replies (P05, P06), include boundary markers and explicit constraints to ensure the agent follows its primary instructions and ignores potential injection attempts within the user-provided data.

ai-prompts-toolkit