THREAT OPS › Threat News › Prompt Engineering for Security Agents: A Measurable Approach with GEPA
Prompt Engineering for Security Agents: A Measurable Approach with GEPA
<p class="wp-block-paragraph">You may have read about our <a href="https://specterops.io/blog/2026/06/09/introducing-ghostworks-ai-cybersecurity-research/">new GhostWorks initiative here at SpecterOps</a>. As part of this effort, we continually trial different methods of evaluating and improving model performance to help understand what techniques can be applied to our research. This post follows
MITRE ATT&CK techniques
Indicators of compromise
- https://gepa-ai.github.io/gepa/tutorials/3d_unicorn_optimization/url
- https://arxiv.org/abs/2507.19457url
- https://gepa-ai.github.io/gepa/guides/#3-optimize_anything-apiurl
- http://body-count:80url
- http://cascade:9999url
- http://ccc:9999url
- http://file-library:9999url
- http://mr-rami:9999url
- http://oreo:9999url
- http://secure-portal:9999url
- http://confused-deputy:9999url
- http://usual-suspects:9999url
- http://warm-up:80url
- http://host:port’url