# Securie

> Agent identity and runtime security for AI coding agents.

Securie is building AgentGuard, a pre-alpha secure runtime for AI coding agents. The platform direction is agent identity, scoped permissions, runtime enforcement, data protection, approval workflows, audit logs, monitoring, and revocation. The current repository is pre-product and contains a CLI scaffold only; enforcement is planned in the vertical slice.

## Core Pages

- [Home](https://securie.io/): Agent identity and runtime security for AI coding agents.
- [Secure Runtime for AI Coding Agents](https://securie.io/secure-runtime-ai-coding-agents.html): Runtime controls for files, commands, network, Git, secrets, approvals, and audit.
- [Agent Identity and Runtime Security](https://securie.io/agent-identity-runtime-security.html): Agent IDs, owners, sessions, permissions, risk tiers, lifecycle, and revocation.
- [AI Agent Policy Engine](https://securie.io/ai-agent-policy-engine.html): Deterministic allow, deny, redact, approval, alert, terminate, and log-only decisions.
- [AI Agent Audit Logs and Compliance](https://securie.io/ai-agent-audit-logs-compliance.html): Evidence for agent activity, policy decisions, approvals, redactions, PR reports, and SOC 2 readiness.
- [Threat Model](https://securie.io/threat-model.html): V1 target actor, escape vectors, enforcement layers, and explicit non-goals.
- [AI Coding Agent Security Checklist](https://securie.io/ai-coding-agent-security-checklist.html): Practical checklist for teams adopting AI coding agents.
- [Claude Code Security](https://securie.io/claude-code-security.html): Guardrails for Claude Code workflows using hooks, runtime controls, policy, and audit.
- [OpenAI Codex Security](https://securie.io/codex-security.html): Runtime controls for Codex-style coding agents.
- [Cursor AI Security](https://securie.io/cursor-ai-security.html): Repository, secret, terminal, Git, and audit controls for Cursor-style workflows.
- [AI Agent Sandbox Alternative](https://securie.io/ai-agent-sandbox-alternative.html): Why sandboxing should be paired with identity, policy, approvals, audit, and revocation.

## Product Status

- Status: pre-alpha, pre-product.
- Current implementation: `agentguard run -- <command>` passthrough only.
- Planned first slice: block `.env` reads, block destructive commands, allow normal repository reads/writes, redact secrets in audit logs, and render session reports.
- Important limit: V1 targets honest hallucinating agents, not malicious operators with kernel-bypass intent.