faultseed
九个确定性钩子，阻止编码代理削弱测试、吞没错误或绕过类型检查——每个都由一个预先植入失败的测试来证明该防护确实能触发。支持 Claude Code 和 DeepSeek Harness。
Nine deterministic hooks that block a coding agent from weakening tests, swallowing errors, or stubbing type checks — each backed by a planted-failure test proving the guard can actually fire. Runs on Claude Code and DeepSeek Harness.
怎么安装
dsh plugin add github:JW53222/faultseed git clone https://github.com/JW53222/faultseed ~/.claude/skills/faultseed 简介
faultseed Deterministic Claude Code hooks that stop a coding agent from faking progress: weakening a test, swallowing an error, deleting a test through the shell, declaring a method that only exists to the type checker. Nine guards ship in this pack. Each one is a subprocess that reads one JSON event on stdin and answers with its exit code — not a linter you run afterward, not a prompt you hope the agent reads, but a gate that refuses the tool call before it happens. Nine guards, nine wired hooks — the default install wires those and nothing else. One further hook ships but is **not** wired by…
推荐参考
信息
- 协议
- MIT
- 语言
- Python
- GitHub 星标
- 1
- 月下载
- –
- 最近更新
- 2026-08-20
- 创建于
- 2026-08-14
基础安全检查
- 检查结果
- curated 收录但无 npm 包/安装命令
- 收录来源
- curated:awesome-dsh-plugin/awesome-dsh-plugin
- 主题标签
- agent-safety, agent-testing, ai-agents, claude-code, cordis, deepseek-harness, dsh, dsh-plugin, fault-seeding, guardrails, hooks, test-integrity
同类推荐
langfuse
langfuse/langfuse
🪢 开源AI工程平台：LLM评估、可观测性、指标、提示管理、游乐场、数据集。与OpenTelemetry、LangChain、OpenAI SDK、LiteLLM等集成。🍊YC W23