Vendor-neutral local framework and CLI for defining, validating, and synchronizing engineering intent across AI coding tools. Offline-first and non-destructive.

Framework-free <ag-ui-chat> Web Component over the AG-UI protocol. Drop-in chat sidebar with a pluggable client-side tool registry, DOM driver primitives, animations, and destructive-action confirmation modal.

Security plugin for OpenClaw - secret detection, PII filtering, and destructive command protection

Agent security framework. Immutable audit trail, permission policies, self-modification detection, destructive command guard.

Human-in-the-loop approval middleware for OpenClaw. Intercepts tool execution via before_tool_call and routes destructive calls through the built-in approval flow (TUI / Control UI / chat).

Scaffold the anpunkit AI-coding workflow (Claude Code) into a project. Upgrade-safe, non-destructive.

LLM security guardrails for OpenCode — jailbreak detection, prompt-injection and web-content scanning, and destructive-command authorization. Bundles enforcing hooks plus the AgentGuards MCP server.

Zero-dependency sandbox wrapper to run CLI agent commands safely, blocking destructive patterns and enforcing execution limits.

Deterministic destructive-command guard + inviolable-constraint policy engine and deviation state machine for the Pi coding agent

Runtime that enforces the OpenWar framework: phase-aware agent loop with confirmation, blocker, and destructive-action gates.

Built-in read-only EVM tools (EvmCall, EvmGetLogs, EvmGetTransaction, EvmGetBalance, EvmBlockNumber). Slice 0 of §47 — destructive transaction tools land in slice 1 with tool-evm-tx.

Security hooks for Pi to reduce accidental destructive actions and secret-file access.

Bridge between dcg (Destructive Command Guard) and Hermes Agent — pre-command scanning, cron-safe reports, workspace protection for AI agent workflows

Belay — a local-first guardrail for AI coding agents (a Lunara product). Intercepts an agent's database/infra actions, computes the blast radius, blocks destructive ops until keychain-gated human approval, and writes a cryptographically signed audit log.

Scan any MCP server for prompt injection, tool poisoning, leaked secrets, and destructive capabilities. One command, no key, no signup.

Hook that blocks destructive edits to protected identity files. For Claude Code CLI and OpenClaw.

Dynamic Workflows: runtime subagent guard (before_tool_call priority 11, fail-closed destructive-op blocking for :subagent: sessions) + LLM-driven orchestration skill (.prose generation, 11 patterns, role-prompt templates). Imports permission primitives f

Non-destructive CLI that scans your own HTTP APIs for common security and hygiene problems, mapped to the OWASP API Security Top 10.

ABI → typed-tool generator. Reads contracts[] from the IR; emits read tools (function stateMutability view/pure) and write tools (destructive: true, gated through wallet-engine). §47 slice 1.

AxioRank content-inspection engine: detectors + risk scoring for AI agent tool calls. Catches secrets, prompt injection, PII, destructive ops, egress, malware and crypto/financial identifiers. Pure, universal, zero runtime dependencies.