Security & AI safety

AI guardrails & agent security

3,625 agents · most starred first

AI guardrails & agent security, page 34 of 37

agentwall-sdk

A lightweight safety, approval, audit, replay, and rollback layer for AI agent tools

radlab-llm-router-plugins

Plugins for the LLM Router (guardrails, maskers, etc.)

vouch-agent

Security scanner for AI-agent skills: audit every SKILL.md on your machine for malware, prompt injection, and secret exfiltration. Deterministic static analysis (with an optional LLM layer) for Claude, Cursor, and MCP skills — like npm audit / semgrep, but for agent skills.

agentid-by-agentmail

AgentID by AgentMail review: compare features, pricing, use cases, access model, and alternatives for this Authentication Agents agent in 2026. Passwordless authentication for AI agents using their own email addresses.

handoffprobe

Adversarial security testing for AI agent handoffs

@once-agent/sdk

Execution safety for side-effecting AI agent tools.

yeknal

Discover AI agent skills on demand, manage manual installs, and run static security audits

coderifts

Signed, offline-verifiable authorization for AI-agent contract changes. Only a granted change can proceed.

skill-preflight

Pre-install safety, token, and maintainability scorecard for AI agent skills.

@stroq/cli

Find known secrets in local Claude Code, Codex and Cursor sessions; trace risky AI agent actions and guard future tool calls

moss-signing

MOSS SDK for TypeScript - Cryptographic signing for AI agent outputs

agent-tool-guard

Tenant and permission isolation for LLM agent tool calls: a principal context on AsyncLocalStorage, a guard that scopes every call and audits it, and a test kit that proves no tool escaped.

namespace-guard

Cross-script confusable detection, slug safety, and LLM Denial of Spend defence. Zero dependencies.

mnki-sdk

mnki — the Agent Trust SDK for TypeScript: give an agent an identity (keys + signed requests), guard() any tool call (local or hosted, observe → enforce), verify, delegate, attest, explain decisions, offline verification. Framework adapters as subpaths: m

@ilyautov/cordon

A prompt-injection firewall for AI agents, with no AI inside: your agent reads anything and takes orders only from you. For Claude Code, Codex CLI, Kimi Code, DeepSeek Harness, Gemini CLI, MCP hosts and LangChain.

@agent-custody/receipts

Chain of custody for AI agents: signed, independently verifiable receipts for tool calls. MCP gateway + Cedar policy + Merkle transparency log

@scopebond/framework

Open-source policy guards and signed records for supported tools in Vercel AI SDK, LangGraph/LangChain, and custom agent applications.

@opena2a/aim-sdk

Agent Identity Management SDK for TypeScript/Node.js

@riskaverse/toolgate

Open auto mode for AI agents: a calibrated tool-call firewall powered by TypeSafe's Jev decision model (direct API, OpenRouter Decisions API, or Vercel AI Gateway). Ships as a Claude Code PreToolUse hook.

actionlog

Tamper-Evident Autonomous Agent Action Audit & Governance Logging MCP Server

@stlw/warden

Deterministic policy engine for autonomous agent tool enforcement

dsh-tm-guard

Autonomous agent permission gate for DeepSeek Harness (dsh) with automatic git/Time-Machine snapshots. Safe defaults on install (protect cwd, deny network/system, block sensitive reads). Configure protectedPaths/extraAllowTools to tune access for tools yo

prooflayer-agent-security

Security scanner MCP server for AI coding agents. Prompt injection firewall, package hallucination detection (4.3M+ packages), 1700+ vulnerability rules with AST & taint analysis, LLM-powered semantic code review, auto-fix. For Claude Code, Cursor, Windsu

@strixgov/tool-gateway

Local-first governed tool execution gateway for AI agents. Read/write classification, policy evaluation, signed execution receipts, multi-key rotation with cross-signed chain snapshots, per-capability rate limits, threshold-based escalation, file/webhook

@ursuciprian/reflex

Open-source pre-execution risk gate and prompt injection guard for AI coding agents (Claude Code, Codex CLI, opencode, pi, oh-my-pi, Hermes): local rules, optional TypeSafe Jev (System One) or local Laya judgments, an editable policy, System 2 escalation

@secureai-sdk/sdk

Enterprise AI Security, GrokBot & Agent Runtime Firewall, In-Process Guardrails, Zero-Trust MCP Proxy & Reversible PII Vault SDK by AcadmyAI

@assistant-ui/passport

Declarative credential verification: JWT, static token, and RFC 7662 introspection providers

@evoclock/pi-agentic-driver

Guardrail extensions for Agentic Driver: advisory review, bounded Herdr communication, and guarded worker lifecycle.

@agenticprimitives/agentic-authorization

Native Agentic authorization discovery: the /.well-known/agentic-authorization document builder, the io.agentictrustlabs.authority MCP extension carrier (read/write), and version negotiation. Public on-chain-derivable config only; not an OAuth authorizati

@emilia-protocol/langgraph

Bind LangGraph Agent Inbox human responses to exact, one-time EMILIA authority. Accept uses the frozen interrupt, edit requires fresh authority for the edited action, and response or ignore never authorize tool execution.

@pqc-sdk/langchain

LangChain / LangGraph structured tool wrappers for @pqc-sdk/core post-quantum cryptographic operations

@guardbee/mcp-agent-graph-auditor

MCP server that builds a reachability graph across multi-agent orchestration configs (LangGraph, CrewAI, AutoGen/ag2) to find transitive excessive agency — an agent with no dangerous tool of its own that can still reach one through delegation or group mem

agent-alibi

We check whether your agent has an alibi for what it did.

agent-assurance

Transparent, framework-agnostic assurance checks for AI agents, mapped to OWASP Agentic security standards.

agent-audit-ledger

Dependency-free tools for an AI agent that keeps records about itself: an append-only ledger verifier, a sandbox egress prober, and an approval-effect checker.

agent-autoguard

A real-time tool-call firewall for AI agents, built on TypeSafe's Jev.

agent-blackbox-jep

Alpha recorder for signed JEP events, evidence digests and incident review

agent-delegation-scope-chain-verifier-mcp

Licensed local MCP server. Verify supplied agent delegation chains, effective scope, action permissions and spend facts before relying on retrospective evidence.

agent-injection-bench

An open benchmark for prompt injection against tool-calling agents.

agent-interlock

Corroboration engine for agent tool-call arguments

agent-mandates

Who authorized an AI agent to act, whether it stayed inside those bounds, and what it cost when it did not.

agent-operating-substrate

Autonomous behavior firewall, vendor-agnostic control plane, and permanent memory across Gemini, Claude, Codex, Copilot, Cursor, and Windsurf

agent-police

Audit LLM API routers for tool-call tampering, credential exfiltration and supply-chain risk

agent-police-gate

Client-side tool-call gate for coding agents: screens what Claude Code and Codex are about to run, and keeps a tamper-evident record of it.

agent-prose

Deterministic Anti-RLHF Stylometric Gate for Autonomous Agents by Nitivra

agent-supply-chain-guard

A conservative scanner for AI-agent skills, MCP manifests, and configurations

agent-sycophancy

Deterministic Epistemic Sycophancy Gate for Autonomous Agents by Nitivra

agent-trust-plane

Authorization boundary and security regression testing for AI agent tool calls (MCP): decide on arguments against delegated authority, bind decisions to single-use grants, replay, and pin denials as CI tests. Experimental.

agentbadge

Agent Badge Python SDK — HTTP client for the Agent Badge API

agentbadge-cli

Agent Badge CLI — enroll, patch, verify, jwks commands

agentbadge-core

Agent Badge canonical business logic: crypto, schemas, credentials, revocation, AgentCard patching

agentccx

AgentScan —— AI Agent 安全检测平台:真实探测区分 AGENT 与裸模型,托管沙箱工具做注入/泄露/越权/外带(DNS+HTTP 金丝雀)/MCP 投毒多维检测,输出带真实证据的报告(支持本地与 docker 目标)

agentconfigsafe

KMS-backed configuration storage for AI agent applications

agentcontainment

Controller-owned authorization and independently verifiable containment for AI agents

agentcrash

Crash-test your AI agent in a synthetic workplace. An open-source agent security test harness with synthetic tools, visual evidence, and regression tests.

agentfox

Agent-native, vendor-neutral governance, security and compliance control plane for AI agents in production.

agentgate-x402

Python SDK for AgentGate x402 — web scraping, prompt injection guard, domain intel. Auto-pays via USDC micropayments on Base L2 / Solana.

agentguard-governance-sdk

Developer SDK for AgentGuard runtime governance and policy enforcement layer

agentguard-shield

Production-quality Python SDK for integrating AI agents with AgentGuard.

agentjail
agentposture

Find every AI agent in your organization and know how risky each one is.

agentpromptsentinel

A multi-tiered defensive pipeline for stopping prompt injection, jailbreaks, and domain hijacking.

agentseatbelt

A seatbelt for AI agents: budget caps, rate limits, time limits, loop detection, endpoint filters and x402 payment limits for any Python agent.

agentsfence

Task-scoped authorization for AI agents: compile user intent into a least-authority policy and enforce it at every tool call. Hermes Agent plugin.

agentshield-os

Deterministic Security, Guardrails, and Governance OS for Autonomous AI Agents in Python.

aiagentallowlist

Python client for the AI Agent Allowlist API: per-URL allow/deny verdicts and verified page-type URLs (login, signup, checkout, upload...) across 40M+ domains for web-browsing AI agents.

aiagentscity-ledger

Spending limits for AI agents: meter every call, cap the spend, and stop it before the provider is contacted.

bayesian-llm-guard

Epistemic Uncertainty Estimation & Guardrails for Agentic LLMs and RAG Systems

boundedllm

Bound what a compromised LLM agent can reach and do

ctrlrun-langchain

Gate a LangChain agent's tool calls with a CTRLRun policy, through wrap_tool_call.

decionis-langchain

Gate LangChain tool calls and LangGraph nodes on a signed Decionis Decision Dossier.

holmesgpt-bridge

This package name has been claimed by Ledger Donjon for security purposes.

interlock-agentgov

Runtime spend governor and denial-of-wallet circuit breaker for autonomous agent fleets.

invaris-agentsec

Adversarial security and reliability testing for autonomous AI agents.

kavros-agent

Kavros governed-egress hook for Python agents (name reservation — the real package ships at kavros.ai launch)

kevlar-agent

Reliability primitives for AI agents that have to survive real use: dedupe, budget guards, silent-timeout, audit trails, confirmation gates, and a self-supervising watchdog.

langchain-clearsigned

LangChain tool for Clearsigned: calibrated claim verification for AI agents, paid per call over x402, receipts verified offline.

langchain-mnki

LangChain / LangGraph integration for mnki (Agent Trust): guard tools so every call is verified against delegated authority, with evidence, before it runs.

langchain-promptfirewall

Sub-millisecond PII detection and prompt injection firewall for LangChain

langgraph-guardrail

Policy enforcement and guardrails for LangGraph AI agents

llama-index-callbacks-promptfirewall

Sub-millisecond PII detection and prompt injection firewall for LlamaIndex

llm-safety-mcp

MCP server for LLM safety checks (prompt injection, PII detection, sanitization)

maigp-claude-agent-sdk

AIGP governance adapter for the Claude Agent SDK (agent-loop level: query/ClaudeSDKClient, tool use). Distinct from aigp-anthropic, which wraps raw Messages API calls.

maigp-microsoft-agent-framework

AIGP governance adapter for Microsoft Agent Framework (MAF) — the converged Semantic Kernel + AutoGen SDK (GA 2026-04-03)

maigp-openai-agents

AIGP governance adapter for the OpenAI Agents SDK (agent-loop level: Runner.run, tools, handoffs). Distinct from aigp-openai, which wraps raw chat completions.

maigp-semantic-kernel

AIGP governance adapter for semantic-kernel

neuraltrust-haystack

NeuralTrust TrustGuard security components for Haystack pipelines

nodra-agent-sdk

Python SDK for protecting autonomous AI-agent actions with Nodra.

once-agent-sdk

Execution safety for side-effecting AI agent tools.

primarypath-claude-agent

Govern Claude Agent SDK tool use through Primary Path (canUseTool gate)

primarypath-langgraph

Govern LangGraph interrupt()-based HITL through Primary Path

primarypath-openai-agents

Govern OpenAI Agents SDK tool approvals through Primary Path

pushary-haystack

Phone approvals for Haystack agents. Your agent asks, your user taps Approve or Deny.

pushary-pydantic-ai

Phone approvals for Pydantic AI agents. Your agent asks, your user taps Approve or Deny.

python-agent-guardrails

A minimal three-layer guardrails template for AI coding agents in Python projects.

reeflex-litellm

Reeflex governance seat for an LLM gateway: a LiteLLM post-call guardrail that rules on every tool call the models behind the proxy propose (allow/deny/require_approval), fail-closed -- one integration for every agent behind the gateway.

sqube-agent-guard

Open-source execution authorization and control infrastructure for AI agents

surf-agentic-base

The record of what an AI agent did, for whom and with whose approval; plus the library a tool server needs.

swarmauth

OAuth 2.1 for autonomous AI agent swarms: signed, short-lived, capability-scoped delegation tokens.

tripwire-llm

Real-time safety supervisor for LLMs — flag a response or abort a stream mid-generation, powered by TypeSafe's Jev.

More in Security & AI safety