Skip to content

v0.4.0 - Prompt injection eval fixture pack

Choose a tag to compare

View all tags
@lindixu6-hash lindixu6-hash released this 16 Aug 20:43
v0.4.0
243cbb5

A framework-neutral, bilingual fixture pack for testing direct and indirect prompt injection boundaries.

Included

  • 8 JSONL cases covering direct override, retrieved web content, email/document injection, tool output injection, data-exfiltration requests, and benign controls
  • required fields for trusted instruction, untrusted content, expected actions, forbidden actions, and evidence
  • machine-readable JSON Schema
  • zero-dependency agentic-validate-fixtures CLI
  • English and Chinese usage guides
  • Coding and Research Agent Cards reference the pack without claiming runtime integration

Safety boundary

Fixtures use synthetic identifiers and reserved .example domains. They contain no live secrets, working collection endpoints, exploit infrastructure, or payload execution.

Use

npm exec --yes \
  --package=github:lindixu6-hash/awesome-agentic-engineering#v0 \
  -- agentic-validate-fixtures path/to/fixtures.jsonl

Verified

  • 57 automated tests pass
  • GitHub-hosted CI runs the validator explicitly
  • clean install from public commit validates all 8 fixtures
  • remote Git tree records the new CLI as executable (100755)

Docs: English · 中文

中文：新增 8 个框架无关的提示注入评估用例、JSON Schema、零依赖校验 CLI 和双语文档。数据集仅使用合成标识符与保留域名，不包含真实密钥、可用攻击端点或 Payload 执行。

Assets 2
Loading