MkaliezZ / dsh-secret-guard

已收录

Fail-closed secret-like payload guard for DSH tool calls: inspects model tool arguments at tools/pre-execute.

main模型工具 查看源代码

安装

npx -y @deepseek-ai/dsh plugin --profile web add github:MkaliezZ/dsh-secret-guard

此安装命令根据 GitHub 仓库地址生成,是未经验证的安装起点。

README

维护者编写的文档快照。

在 GitHub 查看 ↗
提交版本 139172c同步于 2026年8月18日

dsh-secret-guard

A fail-closed secret-like payload guard for DeepSeek Harness (DSH) tool calls.

The plugin inspects model-supplied tool arguments at tools/pre-execute and returns allow, ask, or deny before the tool body runs. v0.1 deliberately does not rewrite/redact arguments: DSH freezes execution identity and arguments before policy, so mutating them inside a guard would violate the runtime contract.

v0.1

  • detects common private-key/token/key patterns;
  • detects obvious sensitive field names;
  • bounds serialized payload size and fails closed above the configured limit;
  • optional tool-name scope (protectedTools); empty means inspect every tool;
  • actionOnFinding: block|ask;
  • no raw secret values in the denial reason.

Non-claims

  • heuristic secret detection, not complete DLP;
  • no guarantee every credential format is detected;
  • no argument redaction in v0.1;
  • not a sandbox or malware detector;
  • should be composed with a policy boundary such as AgentFuse when broader action authorization is required.

Development

npm test

License

MIT

项目文件与信号

以下项目是目录快照中检测到的公开仓库信号。

测试已检测

仓库信息

开发语言
TypeScript
许可证
MIT
最后更新
2026年8月15日 06:35

谨慎安装

请检查源代码、权限、生命周期脚本、依赖与网络访问;不受信任的插件应先在隔离环境中测试。