agentskillexchange Preflight agent specs for prompt-injection risk across prompt, tool, and architecture layers with Prompt Hardener
Describe an agent in agent_spec.yaml, run deterministic prompt-injection analysis, generate mitigations, and validate defenses before rollout.
Prerequisites
Python 3, pipx or uv optional
Installation
Use the upstream install or setup path that matches your environment:
- pipx install https://github.com/cybozu/prompt-hardener/releases/download/v0.6.0/prompt_hardener-0.6.0-py3-none-any.whl
- Recommended if you already use uv for Python tooling.
- uv tool install https://github.com/cybozu/prompt-hardener/releases/download/v0.6.0/prompt_hardener-0.6.0-py3-none-any.whl
- pip install https://github.com/cybozu/prompt-hardener/releases/download/v0.6.0/prompt_hardener-0.6.0-py3-none-any.whl
Requirements and caveats from upstream:
- Deterministic first: init, validate, analyze, report, and diff do not require an LLM API key
- Use this if you prefer a standard Python environment.
Basic usage or getting-started notes:
run deterministic security analysis across prompt, tool, and architecture layers
Choose the installation method that fits how you want to use Prompt Hardener.
Using pipx
Extracted from upstream docs: https://raw.githubusercontent.com/cybozu/prompt-hardener/HEAD/README.md