49 installable, deterministic gates (hooks) for Claude Code: block destructive commands and protected paths, enforce delegation/spec/quality/research rules, and track tasks that only close with verified evidence. Configurable per project.

Destructive changes support for sfdx force:source:deploy

Click / Type / Key / Scroll tools — destructive: true (Section 25 BROW)

BirdyBeep adapter for Claude Code: hook templates, non-destructive config install/uninstall, and event normalization.

Built-in destructive EVM transaction tools (EvmSendTransaction, EvmSimulate). All destructive: true — gated by permission-engine approval + wallet-engine policy. §47 slice 1.

Guard Pi shell commands with Destructive Command Guard.

Agent harness kit for Claude Code — installs verification sensors, destructive-command guards, checkpoint recovery, observability logs, and ratchet skills into your project

Structured, safe Git tool family for DeepSeek Harness: status/diff/log/branch/stage/commit/stash/show/fetch/pull/remote/checkout with a destructive-command guard.

A secure control layer for AI agents: policies across payments, messages, logins, destructive actions, model usage, and secrets, all enforceable from the CLI

Keep the YOLO workflow but avoid bash catastrophe with configurable guardrails for destructive commands.

Wix REST API plugin for OpenClaw — manage blog, CMS data, forms, bookings, contacts, events, FAQ on a single Wix site with site_id whitelist and approval gating on destructive ops

Audit tool arguments and preflight text for destructive commands, sensitive paths, credentials, and remote exfiltration. Advisory only: reports risk, does not block execution.

MCP server for AI agents to evaluate consequences before destructive actions. Analyzes Terraform plans, shell commands, and MCP tool calls.

OpenCode plugin that detects and requires user permission for destructive commands before execution

looprun core: the two-card engine. An AgentSpec says how one desk behaves, a DomainContract says what the business is, and a world card declares its records under reads, writes and destructive. The engine holds every destructive call for consent, states w

DangerZoneSection + ConfirmDialog for destructive actions (delete account, reset, revoke)

Audit each loaded Skill entry Markdown file with bounded heuristics for instruction override, secret exfiltration, destructive commands, and obfuscation; it reports risk labels and does not disable loaded Skills.

Path Guard for pi — blocks destructive commands & path overwrites, protects .env/keys, with strict/normal/loose/trusted/naked guard modes (/guard). pi 防误删/防误覆盖扩展，支持 5 种防护模式。

Terminal proxy that intercepts AI agent shell commands and requires human confirmation before destructive operations

A react component that provides non-destructive inner looping of a video source