AgentDojo vs Crucible
AgentDojo
agentdojo.spylab.aiCrucible
crucible-security.github.ioSide by side
| Figure | AgentDojo | Crucible |
|---|---|---|
| Price from | — | Free |
| Free plan | — | Yes |
| Free trial | — | — |
| Open source | — | — |
| Linux | Yes | Yes |
| API | Yes | — |
| Self-hosted | Yes | Yes |
| Plans published | 0 | 3 |
| Spec score · AI Security Testing Tools | 7.8 (#12) | 7.8 (#14) |
| License | The project's GitHub page identifies its license as MIT. | The core engine and test modules are described as Apache 2.0 and free, with no feature gating on security tests. |
| Purpose | AgentDojo is a dynamic environment for evaluating prompt injection attacks and defenses for LLM agents. | Crucible stress-tests AI agents against attack vectors including prompt injection, goal hijacking, and tool misuse before production. |
Plans
In their own words
AgentDojo
AgentDojo has no write-up on Specifiction yet; its figures are above.
Crucible
Crucible is a self-hosted security scanner for testing AI agents before production. It checks attack behavior such as prompt injection, goal hijacking, jailbreaks, unsafe output, and tool misuse. Instead of reviewing generated text alone, it evaluates tool calls, side effects, and outcomes. The site lists 11 modules and 153 attack vectors, covering eight of ten OWASP Agentic risks, with two in development. Crucible runs 90 attack payloads in under 60 seconds and produces an A–F report with severity, OWASP mapping, and reproduction steps. It supports LangChain, CrewAI, AutoGen, raw MCP endpoints, and agents with HTTP or MCP interfaces. The pip-installed CLI supports Python 3.10+. Local scanning needs no signup and has no telemetry by default. Community is free; Team and Enterprise add monitoring, dashboards, support, SSO, audit logs, deployment options, and custom modules.