AI guardrails & agent security
3,625 agents · most starred first
AI guardrails & agent security, page 35 of 37
Verify an AgentCert + TrustScore inside a Python MCP server (shadow mode, opt-in, dependency-free).
LLM API security auditor — prompt injection, jailbreak, ASCII smuggling, data exfiltration, OWASP LLM Top 10 mapping
Deterministic guardrail, SIEM audit exporter, and subprocess firewall for autonomous AI agents
Volcengine llmscan SDK for Python
Volcengine llmshield SDK for Python
A fail-closed gateway between an AI agent and its tools: normalize, default-deny policy, human approval, append-only audit.
A least-privilege controller for local coding agents
Wellmanifest unified standards runtime — policy-as-code checker for all wellmanifest standards
Python SDK for xAgentGuard AI Agent Governance and Enforcement Proxy
Security and authentication foundation of the ygo74 agent runtime
Runtime for accountable AI agents: signed receipts for every action, human gates, injection defenses, hiring per proven task. Standard library only.
ZTAgent: Zero Trust Security for AI Agents. Never trust an agent action. Verify before execution.
AgentID gives AI agents their own sign-in identity. They can access supported apps without using your password or asking you to log in each time.
Double Agent detection engine: classifies a browser session as human, bot or AI agent, entirely client-side.
Runtime control plane and SaaS governance layer for AI agent tool execution
hush — keep API keys out of AI agent chats. Vault + redacting runner + harness-agnostic MCP server.
MCP idempotency and AI agent retry safety for consequential writes and ambiguous timeouts.
Cryptographic admission and evidence for consequential AI-agent actions.
The Provenance Protocol — an open standard for AI agent identity — with offline verification, signing and attestations
NOA Mandate Core — open, offline-verifiable receipts for AI-agent actions.
AI agent skill security scanner. Detects locally installed agent skills, computes SHA-256 checksums, and queries the Mondoo AI Agent Security database for known threats.
Security policy enforcement and auditing for supported AI agent filesystem tools.
Security and observability for AI agent execution.
Policy enforcement for LLM agent tool calls, decided from each call's message ancestry
MCP server that scans LLM/agent application code for Unbounded Consumption ("denial of wallet") anti-patterns from OWASP LLM Top 10 2025's LLM10 category: missing output token limits, missing timeouts on hand-rolled LLM HTTP calls, unbounded agent tool-ca
LangChain adapter for Proof-Carrying Authority: wrap a StructuredTool so every invocation emits a PCActn (and attaches the proof) before the tool runs.
Safety checks for LangChain.js and LangGraph.js agents that pay: tools to check a transaction or signature before signing, a token before buying, a wallet's open approvals, and an x402 endpoint before paying it (presign-guard and x402 Doctor, paid per cal
Document mutation, read-before-write and admission policies for LangChain agents built on LangGraph.
LangChain.js middleware that verifies Trustflow domain signatures before parsing llms.txt context.
VITNA for agent code: check every tool call before it runs. Adapters for LangChain/LangGraph, the Vercel AI SDK 7, the OpenAI Agents SDK and Mastra. Calls you mark as risky wait for a person. Fails closed.
Overwing SDK: guardrails for LLM output, Atlas user-agent lookups (no key needed), Beacon agent-reachability checks, Preflight checks of a Solana transaction before an agent signs it, and Tower clearance for agents operating legacy systems. Typed client,
See and block what Claude Code, Cursor, OpenAI Codex and GitHub Copilot do. Monitor or Block rules and a signed record for each action. Open source (Apache-2.0); works without an account.
OpenAI Agents SDK (@openai/agents) guard for PCA: guardrail + tool-execution hook so each agent tool call is proof-carrying + policy-gated.
Governed, per-session identity and brokered egress for AI agents
100% Pure Dynamic Behavioral Intelligence Firewall, Speculative Micro-Detonation Sandbox & 1-Terminal Runtime Guard for Autonomous AI Coding Agents
AEGIS / KavachOS — Agentic runtime governance: budget enforcement, DAN gate (human-in-loop), seccomp-bpf kernel enforcement, Falco anomaly detection + Agentic Control Center (cockpit page rendering all @rocketlang/* primitive events in one view, PRAMANA M
Guardrails for the agents you build: permissions, verification, approvals, grounding and budgets from Scute, on the agent framework you already use.
Pluggable envelope encryption + signers + HMAC providers (local-AES / AWS KMS / GCP KMS / AKCS agentic-kms). Pure crypto primitives; no session lifecycle (that's in delegation).
Agentic Primitives Edge admission contracts: transport evidence, application authentication, canonical actor resolution, admission grants, and exact MCP invocation authorization. Zero runtime dependencies; admission is never authority.
AI-agent inventory and tamper-evident action-log verification: auditor-ready traceability evidence for EU AI Act Art. 12
See what your AI agents can cause, route them safely, and prove what happened. Consequence preview, trust routing and signed execution receipts for AI agents.
BRYDGE's verification client: ask before an agent acts, then check against the destination's own records whether the work happened. The shared core every BRYDGE framework adapter is built on.
Ask Lucidrail before a consequential tool call: authorize, wait for a person, report the outcome.
Out-of-band agent governance wrapper. Hard budget caps, eval plateau stop, contract enforcement, and cryptographic receipts.
Out-of-band agent governance wrapper. Hard budget caps, eval plateau stop, contract enforcement, and cryptographic receipts.
Put any MCP-based AI agent (Claude Desktop, Cursor, ChatGPT, LangGraph, ...) under MetaMynd governance without code.
Lightweight client SDK for AegisCore deterministic agent governance (MCP, OpenAI, LangGraph)
Approvals, spend limits and signed receipts for AI agents: ask before a tool pays, shares data or acts, wait for a person, settle, verify. Guards for the OpenAI Agents SDK, LangChain, LangGraph and the Vercel AI SDK; x402 and wallet payments; an LLM gatew
ACGS rule-based tool-call governance for the Haystack Agent before_tool hook
User access reviews for AI agents and MCP servers, with audit evidence and OSCAL output
九维认知空间 Agent 审计层 MCP 插件:决策步骤→九位平衡三进制状态序列→审计报告 (Audit AI agent decisions as balanced-ternary cognitive states via MCP)
Agent Auth RS 校验 SDK(spec 010 P1b):MCP 资源服务器侧 access token 校验 — aud 强校验、命名空间 sub_type 策略、kid 强制 alg、未知 kid 重取限流+负缓存、WWW-Authenticate 发现头。
Local, best-effort egress observation for coding-agent sessions.
AI Agent 环境安全审计 CLI:供应链投毒 / 密钥暴露 / 端口暴露 / MCP server 一键体检
Check that an artifact has the digest its compliance document, SBOM or attestation names.
A runtime security and governance layer that evaluates AI-agent tool requests before they execute.
ACGS rule-based tool-call governance for Microsoft Agent Framework function tools, via function middleware and tool approvals
Pluggable secure-agent platform for Agent Nebula-compatible agents.
Standards-oriented security SDK for Agent Nebula and Agent Genesis
Independent fail-closed second fuse for AI agents: runtime guard + signed receipts + incident reports
AI control layer: per-role virtual MCPs, session taint, local content inspection
Typed policy rails for agent-tool calls: a policy compiler that turns declarative requirements into executable (Type, Predicate, Evidence, Action) rails, a tamper-evident evidence store, trust-scored governance, and an MCP proxy that enforces it all with zero code changes.
Put an AI agent's tool calls behind an Agent Approval Protocol (AAP) approval provider.
A local, deterministic firewall for AI agent tool calls: Claude Code hooks, an MCP proxy and a Python API, with replayable policies for CI
Python SDK for AgentBuck, a real-time trust enforcement layer for AI agents.
Free, account-free messaging utility for agents over an untrusted public hub.
Preflight security auditor for AI-agent and MCP configuration.
Name reservation. Tamper-evident flight recorder for AI agents — audit ledgers, forensics, compliance evidence.
Layered operator screening and session-hijack resistance for AI-agent platforms (alpha)
Python SDK for the AgentGuard AI Agent Authorization API
Find LLM cost risks in Python agent code - unbounded loops, quadratic history growth - and estimate what a run costs.
Static analysis for reliability, cost, and security bugs in AI agents
Open-source safety layer for AI agent tool calls: policy, risk checks, human approval and tamper-evident audit
Exact-action operator approval with durable execution coordination and optional effect adapters
Security chaos engineering for AI agents and LLM applications.
SECURITY RESEARCH PLACEHOLDER — benign, does nothing. Coordinated vulnerability disclosure (dependency-confusion verification) via HackerOne. Not the real package.
Auditable, intent-aware input intervention for LLM agents
AgentLTL rules for coding agents: the AGENTLTL.yaml language, a rule library, and a guard any harness can call.
Security scanner for AI agents and MCP server configurations
Zero-knowledge credential tools for AI agent frameworks
Python client for the AgentShield Tool Security Gateway
Python SDK for AgentTrafficCop — a runtime circuit breaker and schema shield for multi-agent systems.
An append-only, signed ledger for AI agent delegations. ARD finds agents, A2A talks to them — this records who authorised the work, within what limits, and how it ended.
Developer-first runtime authorization SDK for AI agent tool calls. / 面向 AI Agent 工具调用的开发者优先运行时授权 SDK。
Give an AI agent a verifiable AIDR identity: enroll from code, sign requests, LangChain and LlamaIndex tools.
AEAP Composition Boundary Adapter for the Claude Agent SDK. Thin wrapper that attaches the Runtime Decision Point to agent tool invocations — especially useful for coding agents.
AEAP Composition Boundary Adapter for CrewAI. Thin wrapper that attaches the Runtime Decision Point to CrewAI tool invocations.
AEAP Composition Boundary Adapter for LangChain. Thin wrapper that attaches the Runtime Decision Point to LangChain tool invocations.
AEAP Composition Boundary Adapter for LangGraph. Thin wrapper that attaches the Runtime Decision Point to LangGraph node invocations.
AEAP Composition Boundary Adapter for the Microsoft Agent Framework. Thin wrapper that attaches the Runtime Decision Point to agent middleware and workflow execution.
AEAP Composition Boundary Adapter for the OpenAI Agents SDK. Thin wrapper that attaches the Runtime Decision Point to agent tool invocations.
Akeyless Agentic Runtime Authority (RTA/ARA) for AWS Bedrock AgentCore — secretless DB and cloud actions
Aphasia Agentry - canary-proven breach and attack simulation for AI agents (OWASP LLM Top 10)
Microsoft Agent Framework tool guard for Atlas Proof-Carrying Authority (PCA): make every tool/function call proof-carrying and policy-gated
CrewAI tool guard for Atlas Proof-Carrying Authority (PCA): make every CrewAI tool call proof-carrying and policy-gated
Haystack (deepset) tool guard for Atlas Proof-Carrying Authority (PCA): make every Haystack tool/ToolInvoker call proof-carrying and policy-gated
LangGraph tool/node guard for Atlas Proof-Carrying Authority (PCA): make every graph tool/node call proof-carrying and policy-gated, with interrupt()-based step-up for risky nodes
Proof-Carrying Authority tool guard for Pydantic AI (verify a PCActn before each tool runs)
Auth Your Agent SDK: let AI agents act for a user on websites, with the user's approval on their phone.
Full 5-pillar NTI post-quantum security for AutoGen and Microsoft Agent Framework.