# Research

This directory hosts open-source code, datasets, and artifacts produced by
research work related to AI-Infra-Guard. Each research project lives in its
own subdirectory, with its own README, dependencies, and license notes.

## Projects

- [`SkillJack/`](./SkillJack) — Persistent Skill Backdoors in Self-Evolving Agents.
- [`deepseek-harness-security-assessment/`](./deepseek-harness-security-assessment) — Indirect prompt-injection assessment of DeepSeek Harness.
- [`forge_bench/`](./forge_bench) — FORGE-Bench: a deterministic benchmark for studying Loss of Control in autonomous agents.
- [`roguehandoff20/`](./roguehandoff20) — RogueHandoff-20: an isolated Docker benchmark for susceptibility to inherited unsafe agent trajectories.
- [`pikit/`](./pikit) — pikit: a composable prompt-injection research toolkit with attacks, defenses, indirect-injection channels, and agent-framework integrations.