# Spikee — Prompt Injection Eval Kit (CLI)

> ReversecLabs/spikee is a modular CLI for prompt injection/jailbreak evals; verified 184★ and documents `spikee generate` → `spikee test`.

## Install

Copy the content below into your project:

## Quick Use

```bash
python -m venv .venv && source .venv/bin/activate
pip install spikee
mkdir -p workspace && cd workspace
spikee init
spikee list attacks
spikee generate --seed-folder datasets/seeds-cybersec-2026-01 --format user-input
```

## Intro

ReversecLabs/spikee is a modular CLI for prompt injection/jailbreak evals; verified 184★ and documents `spikee generate` → `spikee test`.

**Best for:** Security-minded teams evaluating LLM apps, RAG systems, and guardrails against injection/jailbreak attacks

**Works with:** Python envs, OpenAI-compatible endpoints by default, optional extras for Bedrock/Azure/Ollama/Groq

**Setup time:** 10-30 minutes

### Key facts (verified)

- GitHub: 184 stars · 41 forks · pushed 2026-05-13.
- License: Apache-2.0 · owner avatar + repo URL verified via GitHub API.
- README-backed entrypoint: `spikee generate --seed-folder ...`.

## Main

- Follow the two-stage loop in README: generate datasets from seed folders, then test a target with consistent judges/providers.

- Keep installs lean: README says the default install targets OpenAI-compatible endpoints; use extras only for providers you need.

- Treat results as evidence: keep JSONL datasets + configs and re-run `generate` + `test` in CI or on release candidates.

### Source-backed notes

- README says Spikee migrated away from LangChain to `any-llm` to reduce dependency bloat, with optional extras like `spikee[all]`.
- README quickstart uses `pip install spikee`, then `spikee init` for a workspace and `spikee list` for modules.
- README describes dataset generation (`spikee generate`) and testing (`spikee test`) as the core workflow stages.

### FAQ

- **Is Spikee only for standalone models?**: No — README includes LLM applications/agents and guardrails as targets.
- **Do I have to install heavyweight SDKs?**: No — README says the default install stays minimal and provider SDKs are optional extras.
- **How do I keep evals repeatable?**: Version your seed folders/datasets and re-run with pinned providers/judges.

## Source & Thanks

> Source: https://github.com/ReversecLabs/spikee
> License: Apache-2.0
> GitHub stars: 184 · forks: 41

---

<!-- ZH -->

## Quick Use

```bash
python -m venv .venv && source .venv/bin/activate
pip install spikee
mkdir -p workspace && cd workspace
spikee init
spikee list attacks
spikee generate --seed-folder datasets/seeds-cybersec-2026-01 --format user-input
```

## Intro

ReversecLabs/spikee 是模块化 CLI 工具箱，用于 prompt injection/jailbreak 测试；已验证 184★，README 采用两阶段流程：先 `spikee generate` 生成数据集，再 `spikee test` 测目标。

**Best for:** 要评估 LLM 应用/RAG/guardrails 抗注入与越狱能力的安全团队

**Works with:** Python 环境；默认支持 OpenAI-compatible endpoint；Bedrock/Azure/Ollama/Groq 等通过可选依赖

**Setup time:** 10-30 minutes

### Key facts (verified)

- GitHub：184 stars · 41 forks；最近更新 2026-05-13。
- 许可证：Apache-2.0；作者头像与仓库链接均已通过 GitHub API 复核。
- README 中可对照的入口命令：`spikee generate --seed-folder ...`。

## Main

- 按 README 的两阶段闭环：先从 seed 生成 dataset，再对目标执行 test，并保持 judge/provider 一致性。

- 安装尽量轻：README 说明默认只覆盖 OpenAI-compatible endpoint；需要 Bedrock/Azure/Ollama/Groq 再按需装 extras。

- 把结果当证据：保留 JSONL dataset + 配置，并在 CI 或发版前重复跑 `generate` + `test`。

### Source-backed notes

- README 说明 Spikee 从 LangChain 迁移到 `any-llm` 以减少依赖膨胀，并提供如 `spikee[all]` 的可选 extras。
- README 的 quickstart：`pip install spikee`，再用 `spikee init` 初始化 workspace，并用 `spikee list` 查看模块。
- README 将核心流程分为生成（`spikee generate`）与测试（`spikee test`）两阶段。

### FAQ

- **Spikee 只能测裸模型吗？**：不是。README 明确包含 LLM 应用/agent 与 guardrails 作为测试目标。
- **一定要装很重的 SDK 吗？**：不一定。README 说明默认安装尽量精简，重依赖通过 extras 按需安装。
- **如何保证评测可复现？**：把 seed/dataset 纳入版本控制，固定 provider/judge 后重复跑。

## Source & Thanks

> Source: https://github.com/ReversecLabs/spikee
> License: Apache-2.0
> GitHub stars: 184 · forks: 41


---
Source: https://tokrepo.com/en/workflows/spikee-prompt-injection-eval-kit-cli
Author: Script Depot