AgentGuard plugin
AgentGuard checks a coding agent's actions before they run and scans third-party skills for secrets, backdoors and injections.
- 493 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Hooks intercept every agent action and the optional cloud receives policy and a redacted audit log.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/agentguardDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Hooks intercept every agent action and the optional cloud receives policy and a redacted audit log.
This is third-party code. Review the repository files before installing.
What it does
AgentGuard is a security layer from GoPlus Security for coding agents. `agentguard init` detects installed agents (Claude Code, Codex, OpenClaw, Hermes and others) and wires in hooks. The hooks block `rm -rf /`, `curl | bash`, writes to `.env` and `.ssh/`, and data sent to Discord, Telegram and Slack webhooks, and they show which skill started an action. `agentguard scan` checks a local skill or a GitHub repository before install against 24 rules: secrets, backdoors, obfuscation, prompt injection and, for Web3, wallet draining and unlimited approvals. A trust registry limits what each skill may do. The local runtime guard needs no cloud account.
Who it is for. Developers and teams installing third-party skills and plugins into Claude Code or Codex who want automatic checks before anything runs.
Good fit when
- Before installing a skill from GitHub you want its repository checked for secrets and malicious code.
- The agent works in a directory with `.env` and keys, and dangerous commands must be blocked before they run.
- A team runs a Web3 project and no skill may reach the wallet.
- A skill update is checked for new permissions through a version comparison.
Not a fit when
- You need a full sandbox: the hooks check actions but do not isolate the process.
- The agent runs outside Claude Code, Codex, OpenClaw, Hermes or other supported hosts.
- Static analysis must replace manual review of high-stakes code.
Example request
Scan this skill repository before install and list the risks found, by severity.Limitations
Protection applies only to agents wired up with `agentguard init`; the rules are static, so a hook sometimes blocks a legitimate command and crafty code can pass. The optional `agentguard connect` links the install to AgentGuard Cloud and syncs policy and a redacted audit log. `agentguard subscribe --cron` adds a cron or Windows Task Scheduler job that polls the threat feed. The protection level is set at install time.
How to disable. Remove the hooks and plugin with your agent's own tooling, then run `npm uninstall -g @goplus/agentguard`. If you enabled the subscription, delete the cron job and `~/.agentguard`.
FAQ
How do I vet a skill before installing it?
Run `agentguard scan ./path` for a local folder or `agentguard scan https://github.com/owner/repo --json` for a repository. Pass `--ref` with a full commit SHA for a reproducible report.
Is a cloud account required?
No. The local guard works without the cloud or a network. Cloud is optional via `agentguard connect` for policies and audit sync.
Which agents are supported?
`agentguard init` detects installed agents automatically; choose one with a flag such as `--agent claude-code` or `--agent codex`. The list also covers OpenClaw, Hermes and QClaw.
What if a hook blocks a command I need?
Check the command by hand and adjust the protection level instead of disabling the hooks entirely.
Is there Web3 scanning?
Yes, the rules cover wallet draining, unlimited approvals, reentrancy and proxy exploits.
Discussion
No questions or reviews yet.
Related
PluginsA local guard for coding agents: blocks secret writes and force pushes to main, warns about risky commands.
Security hook for coding agents: scores each tool call with session context and flags injections in results and skills
A CLI that scans your machine for skills and MCP configs, finds poisoned tool descriptions and runs a system prompt through 225 attacks
HOL Guard checks an AI agent's shell commands, secret reads, MCP calls and package installs before they run and holds risky actions until you decide
A skill and local program for agent safety: inventory, skill admission checks, human-approved grants and signed receipts.
An AI security verification CLI: scans models and code, probes live endpoints and MCP servers, and analyzes recorded agent runs
README in short
Three layers: automatic guard hooks, a deep skill scan and a daily patrol. Install via npm and `agentguard init`.