Openmoat — Security skill for Claude Code
Security for AI coding agents.
How to install Openmoat
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open crocodile-labs/openmoat and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Openmoat does
Security for AI coding agents. One policy controls what Claude Code, Codex and Cursor can run, read, write and connect to, enforced by OS sandboxes, with a tamper-evident audit log.
Alternatives in Security
- Anthropic Cybersecurity Skills — 734+ structured cybersecurity skills for AI agents · MITRE ATT&CK mapped · agentskills.io open standard · Work 3.8k ★
- OpenTag — Open-source, channel-native agent gateway for Slack 499 ★
- Loongsuite Pilot — Local-first telemetry collector for AI coding agents — unified OpenTelemetry events for Claude Code, Codex, Cu 150 ★
README
OpenMoat
**Security for AI coding agents.** OpenMoat checks every action an AI coding agent takes on your computer and stops the dangerous ones before they run.
Works with **Claude Code**, **Codex** and **Cursor**. Built by Crocodile Labs. The command is `moat`.
Why you need it
AI coding agents run commands, edit files and connect to the internet with your account and your permissions. Agents also read untrusted text all day: READMEs, issues, web pages, package code. One hidden instruction in that text can make an agent:
- copy your SSH keys, cloud credentials or
.envsecrets to someone else; - delete files, force-push over your branch, or rewrite your shell startup files;
- change its own settings so that it can do more next time.
The agent's built-in "allow this?" prompt shows you a command, not what that command does. A harmless-looking `npm test` can run code that reads `~/.ssh`. And in auto-approve modes, in CI and in background agents, there is no prompt at all.
How it works
Before the agent runs a command, reads or writes a file, opens a web page or calls an MCP tool, OpenMoat checks the action against one policy that you control:
| Decision | What happens |
|---|---|
| Allow | Normal work runs without interruption. |
| Ask | You are asked first, and told why. |
| Deny | The action is blocked. The agent is told which rule matched and why. |
agent action ──► OpenMoat ──► allow ──► runs, and is recorded
├► ask ──► you decide, with the reason shown
└► deny ──► blocked, with the rule and the reason
The same policy also configures the operating system's sandbox for each agent. Code hidden inside an approved command (a test script, a build step) still cannot read your secrets or write outside the project.
What it protects
- Your secrets. SSH keys, cloud credentials, tokens and
.envfiles stay out of reach. The secrets broker gives the agent a placeholder i
Related Skills
Sanctuary Framework
Open-source security for AI agents: kernel-enforced egress control on macOS and Linux, keys only the operator
Trustgate
Deterministic policy enforcement and a tamper-evident audit log for AI agents. Decides what your agent may act
Memnox
See what your AI coding agents can actually do on your machine, and put the dangerous actions behind ask or de
Coding Agent Security Checklist
Stop your AI coding agent (Claude Code, Cursor, Codex) from installing malware straight out of llms.txt docume
ContAIned
A governed coding agent CLI built on Claude Code. Runs inside an isolated Docker container with operator-contr
Guardskill
Read-only scanner for git settings that let a repository run code in coding agents (Claude Code, Codex, Cursor
Related Agents
Security Review Readonly
READ-ONLY security review for Phase 1 pipeline. Finds HIGH-confidence vulnerabilities. Cannot Edit/Write files
Threat Mitigation Mapping
Connect threats to controls for effective security planning. - wshobson/agents
Audit LinkedIn
LinkedIn Ads evidence and controls specialist. Returns schema-valid findings for Insight Tag and conversions,