作者 kenryu42 · kenryu42/claude-code-safety-net
A hooks-driven safety net for Claude Code — interceptors catch rm -rf, git reset --hard, force pushes, and uncommitted destructive edits.
Installs as a Claude Code plugin with PreToolUse hooks. Each potentially destructive operation is intercepted and either confirmed or refused based on rules. Includes rules for git, file system, npm, and arbitrary bash.
何时使用： You've been bitten by an agent overwriting uncommitted changes.
git reset --hard; expect block.✓ 已复制结果： Destructive ops require explicit override.
Pairs naturally with any agent doing repo edits
| 工具 | 输入参数 | 何时调用 | 成本 |
|---|---|---|---|
| hook-driven | (see docs) | No callable tools; activates via PreToolUse hooks | 1 call |
hooks-read hooks-writeHooks not firingConfirm settings.json was actually updated; restart Claude Code
False positives on benign git opsTune the per-project rules file
| 替代方案 | 何时用它替代 | 权衡 |
|---|---|---|
| pegasi-ai/reins | You want broader agent guardrails | Heavier setup |
| Manual review | Solo, low-stakes repo | Doesn't scale |