The Agent Skills Directory

[SAFE]: The skill serves as a comprehensive guide for agent evaluation, covering regression testing, statistical analysis, and behavioral contracts. It promotes security-first practices such as data anonymization and credential protection.
[PROMPT_INJECTION]: The architectural sketches include examples of prompt injection strings (e.g., 'Ignore all previous instructions', 'What are your system instructions?'). These are explicitly documented as part of an adversarial testing methodology designed to assess agent robustness and are not instructions for the agent to execute.
[OBFUSCATION]: A base64 encoding pattern is mentioned in the adversarial testing section. This is used solely to demonstrate how developers can test an agent's ability to resist encoded injection attacks and is not used to hide malicious intent within the skill itself.
[DATA_EXPOSURE]: The skill provides strong guidance on avoiding data exfiltration, specifically instructing users to supply synthetic fixtures instead of real customer records and to ensure that trace retention excludes credentials.
[DYNAMIC_EXECUTION]: While the skill contains TypeScript/JavaScript patterns for statistical evaluation and regression testing, these are presented as optional architecture sketches and non-runnable examples rather than executable code distributed with the skill.

agent-evaluation