A safety hook that blocks known destructive shell commands before they run, including commands that erase files, remove database tables, or force-push code.

Use it to guard coding-agent tool calls against destructive Bash commands such as recursive deletion, database destruction, unsafe download-and-run commands, force-pushes, and device wipes.

It adds a last-minute check against commands that could permanently damage project files, data, devices, or version history.

bouncer — Plugin by karanb192

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

Claude Code

/plugin marketplace add karanb192/bouncer

agentmods

npx agentmods add plugins/karanb192/bouncer/claude

Clone the repo

git clone --depth 1 https://github.com/karanb192/bouncer

Made for: Claude Code.

Per session not measured What this adds to a session before it is invoked.

When invoked not measured Not applicable: nothing here is loaded into a session.

Security scan C 2 findings. Scan, not verified.

Origin original No closer match found in the catalogue.

Security

Grade C, and why

bouncer scanned grade C with 2 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Recursive force deletehighDestructive command

rm -rf with a variable or a broad path is one typo away from removing the wrong tree.

"description": "A one-file PreToolUse door-guard that bounces destructive Bash commands (rm -rf, DROP TABLE, curl|sh, force-push, device wipes) before they run. Blocks 45/45 known footguns, 0 false positives.",

Makes network callslowCapability

Not a fault in itself. Listed so you know the mod talks to something, and to what.

"description": "A one-file PreToolUse door-guard that bounces destructive Bash commands (rm -rf, DROP TABLE, curl|sh, force-push, device wipes) before they run. Blocks 45/45 known footguns, 0 false positives.",

claude/.claude-plugin/plugin.json · 21 lines

What it actually says

{
  "name": "bouncer",
  "version": "1.0.1",
  "description": "A one-file PreToolUse door-guard that bounces destructive Bash commands (rm -rf, DROP TABLE, curl|sh, force-push, device wipes) before they run. Blocks 45/45 known footguns, 0 false positives.",
  "author": {
    "name": "Karan Bansal",
    "url": "https://github.com/karanb192"
  },
  "homepage": "https://github.com/karanb192/bouncer",
  "license": "MIT",
  "keywords": [
    "security",
    "safety",
    "hooks",
    "guardrails",
    "claude-code",
    "pretooluse"
  ],
  "hooks": "./hooks/hooks.json"
}

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

2d ago First seen · 21 lines scan C a38aa8659e71

Subscribe to this mod's changes

bouncer is a plugin published in the GitHub repository karanb192/bouncer (5 stars, last pushed 15d ago), licensed MIT. Its token cost is not measured: this kind of file is read by the harness, not the model. A static security scan graded it C with 2 findings (recursive force delete, makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Databases Guardrails Security ai-agents claude-code developer-tools hooks safety

probity

TDD enforcement and guardrails for Claude Code, Codex, and GitHub Copilot CLI.

nizos/probity · not measured

probity

Plugin marketplace listing 1 plugin: probity.

nizos/probity · not measured

makoto

Plugin marketplace listing 1 plugin: makoto.

Clear-Sights/Makoto · not measured

makoto

Holds Claude Code to its word — every claim checked against the agent's own logged record; faked checks blocked, not warned (four narrow advisory self-checks excepted). Two-tier catalog (15 pre-checks, 19 Stop gates), each shipped at measured zero false positives.

Clear-Sights/Makoto · not measured