Spikee
Simple Prompt Injection Kit for Evaluation and Exploitation
Install / Use
/learn @ReversecLabs/SpikeeREADME
Version: 0.7.1-dev
Developed by Reversec Labs, spikee is a toolkit for assessing the resilience of LLMs, guardrails, and applications against prompt injection and jailbreaking. Spikee's strength is its modular design, which allows for easy customization of every part of the testing process.
Architecture Update: Spikee has migrated away from LangChain in favor of
any-llm. This decision was made to significantly reduce dependency bloat. Now, by default,spikeeinstalls only the bare minimum required to support providers with OpenAI-compatible API endpoints (e.g., OpenAI, DeepSeek, Google, TogetherAI, OpenRouter). Providers for which spikee relies on native SDKs (like AWS Bedrock, Azure, Ollama, or Groq) are supported through lightweight optional dependencies (e.g.spikee[all]), ensuring users only install the heavy SDKs they actually need.
Table of Contents
- Spikee Use Cases
- The Spikee Architecture
- A Quick-Start Guide to Using Spikee
- Contributing
- Questions or Feedback?
Documentation
- Guides
- Reference Pages
- Generation and Testing Workflows
- Custom Modules and Development
- Additional Resources
Spikee Use Cases
Spikee can be used to test
- LLMs in isolation (traditional "LLM red teaming")
- GenAI features within LLM applications/agents (such as chatbots, RAG systems, etc.)
- LLM guardrails
The Spikee Architecture
Spikee operates in two stages: generating a test dataset, and executing tests against a target using the dataset. Each stage is powered by easy-to-customize Python modules.
<div align="center"> <img src="docs/spikee-architecture.png" width="700px"> </div>A Quick-Start Guide to Using Spikee
1. Installation
1.1 Install spikee directly from PyPI.
pip install spikee
Note: To keep the installation lightweight, this command only installs the base dependencies required to connect to OpenAI-compatible API endpoints (which covers OpenAI, DeepSeek, OpenRouter, TogetherAI, Google, etc). If you plan to use providers for which spikee relies on native SDKs (like Bedrock, Azure, Ollama, or Groq), install the necessary extras:
pip install "spikee[all]"
# Or choose specific providers, e.g., "spikee[bedrock,azure,ollama,groq]"
1.2 Local Installation (From Source)
git clone https://github.com/ReversecLabs/spikee.git
cd spikee
python3 -m venv env
source env/bin/activate
pip install ".[all]"
Development Guidance: pip install -e ".[all]" will create a symlink within the venv, allowing you to make changes to the codebase without needing to reinstall after each change.
1.3 Optional Plugin & Target Dependencies
Spikee features several sample plugins and targets that require specific third-party libraries. Instead of cluttering the global installation, these can be installed as needed:
- Local Inference (
torch,transformers,sentencepiece): Required for local models and the OPUS-MT translation plugin. - Google Translate (
googletrans): Required for the Google Translate plugin. - PDF Generation (
fpdf2): Required for the sample PDF target.
pip install "spikee[local-inference]"
pip install "spikee[google-translate]"
pip install "spikee[pdf]"
2. Your Workspace
Spikee requires a workspace to store datasets, results and local modules (targets, plugins, attacks, judges). Create a folder called workspace and run the spikee init command to populate it with built-in datasets and sample modules.
mkdir workspace
cd workspace
spikee init
See spikee init documentation for information on --include flags (e.g., Spikee Viewer).
3. Available Modules: spikee list
Use spikee list to see what seeds, datasets, judges, targets, plugins and attacks are available in your workspace (both local and built-in).
spikee list seeds
spikee list datasets
spikee list judges
spikee list targets
spikee list plugins
spikee list attacks
spikee list providers --description
(NB, use --description to get a brief description of each module - not supported on seeds or datasets).
4. Generating a Dataset: spikee generate
spikee generate is used to create custom datasets from seed folders, including options to apply transformations through plugins and formatting modifiers to tailor the dataset to specific targets. Your testing scenario will determine what datasets you need to generate.
A list of built-in datasets and plugins is available within the Built-in Datasets and Built-in Plugins documentation and a complete list of dataset generation options is available in the Dataset Generation documentation.
4.1. Choosing a Dataset Generation Format
Scenario A: Testing an LLM Application
When testing an application (e.g., chatbot or email summarisation tool), you don't typically control the entire input to the LLM, tha is the system message and/or instructions the application is passing to the LLM. You only control the user input (e.g., chat message or document) which the application passes as a parameter to the prompt templates.
- What to Generate: Just a user prompt or document with the payload.
- How to Generate: Use
--format user-input(you can omit this flag as it is the default).
spikee generate --seed-folder datasets/seeds-cybersec-2026-01 --format user-input
This will generate the dataset in JSONL format: datasets/cybersec-2026-01-document-dataset-<TIMESTAMP>.jsonl.
Scenario B: Testing a Standalone LLM
When testing an LLM directly, you control the entire prompt fed to the LLM. This is ideal for assessing a model's general resilience to jailbreaks and harmful instructions.
- What to Generate: A full prompt, which includes a task (like "Summarize this: <data>"), the data containing the prompt injection or jailbreak, and optionally a system message.
- How to Generate: Use
--format full-promptand optionally--include-system-message. Thedatasets/seeds-cybersec-2026-01folder provides a great starting point with diverse jailbreaks and attack instructions.
spikee generate --seed-folder datasets/seeds-cybersec-2026-01 --format full-prompt
This will generate the dataset in JSONL format: datasets/cybersec-2026-01-full-prompt-dataset-<TIMESTAMP>.jsonl.
4.2. Generating Standalone Attacks
Spikee typically uses composable datasets, that combine permutations of user inputs, jailbreaks and instructions. However, it also supports Standalone Inputs which are ready-to-use prompts, without any additional formatting or composition. This is useful for quickly testing specific prompts, or using publicly sourced datasets that contain ready-to-use attack prompts, such as seeds-simsonsun-high-quality-jailbreaks.
To include those in the generated dataset, we use --include-standalone-inputs:
spikee generate --seed-folder datasets/seeds-simsonsun-high-quality-jailbreaks \
--include-standalone-inputs \
4.3. Transformations using Plugins
Datasets can be enhanced using Plugins, which apply transformations to payloads at the time the dataset is generated (this is in contrast to Attacks, which are applied dynamically at the time of testing). This allows you to assess transformation-based jailbreak techniques.
See Built-in Plugins for a list of available plugins and Creating Custom Plugins for guidance on writing your own.
Usage
# 1337 (leetspeak) plugin
spikee generate --seed-folder datasets/seeds-cybersec-2026-01 \
--plugin 1337
# Best of N plugin, with 50 variants per entry
spikee
