Tools
Security Control
By Openclaw security monitoring and runtime intervention plugin
Install
npm install --legacy-peer-depsREADME
# sst-security-control-openclaw
**English** · [简体中文](README.zh-CN.md)
An open-source runtime security plugin for OpenClaw. It provides local risk detection, security event logging, and blocking of high-risk operations.
**Highlights**: batteries included. Runs locally out of the box — no account registration, no cloud connectivity.
## Installation
```bash
# 1. Build the package
npm install --legacy-peer-deps
npm run build:prod
# 2. Copy the tarball to the target server
scp install/sst-security-control-openclaw-1.0.1.tgz root@server:~/.openclaw/
# 3. Install on the server (run these two commands in separate terminal sessions)
openclaw plugins install npm-pack:$HOME/.openclaw/sst-security-control-openclaw-1.0.1.tgz --accept-capabilities --force
openclaw plugins enable sst-security-control-openclaw
openclaw gateway restart
```
## Uninstall
```bash
openclaw plugins uninstall sst-security-control-openclaw
```
## Capabilities
- Agent session and behavior event logging (NDJSON)
- High-risk actions (reverse shells, sensitive file access, and similar) are blocked in `enforce` mode; everything else is only flagged and warned
- Runtime status and event panel (Control UI → SST Security Control)
## Configuration
The default configuration shipped with a fresh install is already tuned to a production baseline and works as-is. To adjust it:
```bash
openclaw plugins config sst-security-control-openclaw --set \
"intervention.mode=audit" \
"application.enabled=true"
```
Key configuration options:
| Option | Default | Description |
|------|--------|------|
| `intervention.mode` | `enforce` | `audit` (log only) or `enforce` (block) |
| `application.enabled` | `true` | Application event collection switch |
| `application.runtimeHooks.enabled` | `false` | Enable runtime hooks (requires NODE_OPTIONS setup) |
| `intervention.failStrategy` | `fail_closed` | Behavior when the intervention engine fails or times out (`fail_open` allow / `fail_closed` block) |
| `collector.ndjsonPath` | `~/.openclaw/security-events.jsonl` | Event log path |
## Custom SPL Rules
Blocking is driven by SPL declarative rules (YAML, `apiVersion: security.use/v1`). A ready-to-use rule set ships with the package; see [policies/security-policy.spl.yaml](policies/security-policy.spl.yaml) for rule examples. Before switching to `enforce` mode, copy the policy file to `~/.openclaw/security-policy.yaml` and review the rules.
## Security Events
Events are NDJSON lines. The main event types are:
- `agent_session_start` / `agent_start` / `agent_stop`
- `agent_tool_call` / `agent_tool_result`
- `runtime_spawn_intent` / `runtime_spawn_denied`
- `security_audit` (embedded guard hits, loop detection, token anomalies)
## Roadmap
- [x] **1.0.0 · Event collection + observability + rule analysis**: NDJSON event stream, Control UI panel, SPL rule security analysis
- [ ] **1.1.0 · Semantic detection SDK**: introduce a standalone semantic detection SDK alongside the existing SPL rule engine to improve detection coverage
## Project Home
- Website: [https://www.zhihuizhanxing.com](https://www.zhihuizhanxing.com)
- Source: [https://github.com/zhanxingtai/openclaw-security-control](https://github.com/zhanxingtai/openclaw-security-control)
- X: [@StarSightTec](https://x.com/StarSightTec)
- WeChat Official Account:
## Support
To report a security vulnerability or request urgent support, contact us by email (please do not disclose security details in a public GitHub issue):
- Security email: <support@zhihuizhanxing.com>
General usage questions and feature requests go to GitHub Issues.
## License
[Apache-2.0](LICENSE) — dependency declarations are in `package.json`.
tools
Comments
Sign in to leave a comment