# Prompt Hardener — Prompt-Injection Risk Analyzer

> Prompt Hardener analyzes prompt-injection risk from `agent_spec.yaml`, exports reports, and can simulate adversarial attacks with API keys.

## Install

Paste the prompt below into your AI tool:

## Quick Use

```bash
uv tool install https://github.com/cybozu/prompt-hardener/releases/download/v0.6.0/prompt_hardener-0.6.0-py3-none-any.whl
cp examples/chatbot-minimal/agent_spec.yaml ./agent_spec.yaml
prompt-hardener validate agent_spec.yaml
prompt-hardener analyze agent_spec.yaml --format markdown
prompt-hardener report analyze.json -f html -o report.html
```

## Intro

Prompt Hardener analyzes prompt-injection risk from `agent_spec.yaml`, exports reports, and can simulate adversarial attacks with API keys.

**Best for:** agent builders who want a repeatable security checklist for prompts/tools/architecture

**Works with:** Python, pipx/uv, YAML specs, CI pipelines (Markdown/HTML/JSON outputs)

**Setup time:** 8-15 minutes

### Key facts (verified)

- GitHub: 51 stars · 7 forks · pushed 2026-05-12.
- License: Apache-2.0 · owner avatar + repo URL verified via GitHub API.
- README-verified entrypoint: `pipx install https://github.com/cybozu/prompt-hardener/releases/download/v0.6.0/prompt_hardener-0.6.0-py3-none-any.whl`.

## Main

- Start deterministic: `validate` + `analyze` give you a static run without any LLM API key (per README).

- When you want to test defenses, use `simulate` on a hardened spec to run adversarial scenarios (README notes this is LLM-backed).

- Treat the output as CI artifacts: export Markdown/HTML/JSON and diff runs over time to track regressions.

### Source-backed notes

- README describes a single workflow driven by `agent_spec.yaml` and supports exports as Markdown/HTML/JSON.
- README quick start copies an example spec and runs `prompt-hardener validate` and `prompt-hardener analyze` without API keys.
- README lists which commands require API keys (`simulate` yes; many others no).

### FAQ

- **Do I need an API key for analysis?**: No — README says `validate`/`analyze` are deterministic and do not require keys.
- **What file do I write?**: An `agent_spec.yaml` describing your system, as shown in the README examples.
- **Can I export a report?**: Yes — `report` renders JSON results to HTML/Markdown/JSON per README.

## Source & Thanks

> Source: https://github.com/cybozu/prompt-hardener
> License: Apache-2.0
> GitHub stars: 51 · forks: 7

---

<!-- ZH -->

## Quick Use

```bash
uv tool install https://github.com/cybozu/prompt-hardener/releases/download/v0.6.0/prompt_hardener-0.6.0-py3-none-any.whl
cp examples/chatbot-minimal/agent_spec.yaml ./agent_spec.yaml
prompt-hardener validate agent_spec.yaml
prompt-hardener analyze agent_spec.yaml --format markdown
prompt-hardener report analyze.json -f html -o report.html
```

## Intro

Prompt Hardener 通过 `agent_spec.yaml` 分析 prompt 注入风险，支持确定性静态检查与报告导出；配置 API key 后还可运行对抗模拟，验证防护效果。

**Best for:** 想要可复现的 prompt/tool/架构安全检查清单的开发者

**Works with:** Python；pipx/uv；YAML 规格文件；CI（Markdown/HTML/JSON 输出）

**Setup time:** 8-15 minutes

### Key facts (verified)

- GitHub：51 stars · 7 forks；最近更新 2026-05-12。
- 许可证：Apache-2.0；作者头像与仓库链接均已通过 GitHub API 复核。
- README 中核对过的入口命令：`pipx install https://github.com/cybozu/prompt-hardener/releases/download/v0.6.0/prompt_hardener-0.6.0-py3-none-any.whl`。

## Main

- 先走确定性流程：按 README，`validate` + `analyze` 不需要 LLM API key 就能跑完整静态分析。

- 需要验证防护效果时，对 hardened spec 执行 `simulate`（README 说明该步骤需要 API key）。

- 把输出当 CI 产物：导出 Markdown/HTML/JSON，并用 `diff` 追踪安全回归。

### Source-backed notes

- README 说明流程由 `agent_spec.yaml` 驱动，并支持导出 Markdown/HTML/JSON。
- README 的 Quick start 通过复制示例 spec，执行 `validate` 与 `analyze`（无需 API key）。
- README 列出哪些命令需要 API key（如 `simulate`）。

### FAQ

- **做静态分析需要 API key 吗？**：不需要。README 说明 `validate`/`analyze` 为确定性流程。
- **需要写什么文件？**：写 `agent_spec.yaml` 描述系统，README 有示例。
- **能输出报告吗？**：能。README 说明 `report` 可渲染为 HTML/Markdown/JSON。

## Source & Thanks

> Source: https://github.com/cybozu/prompt-hardener
> License: Apache-2.0
> GitHub stars: 51 · forks: 7


---
Source: https://tokrepo.com/en/workflows/prompt-hardener-prompt-injection-risk-analyzer
Author: Prompt Lab