Tools
Wraithvector Openclaw
By Runtime governance plugin for OpenClaw agents — policy enforcement, command scope control, and cryptographic audit trail. Powered by WraithVector.
Configuration Example
{
"exec": {
"allowed_roles": ["*"],
"allowed_commands": ["ls", "pwd", "echo", "cat", "grep"]
},
"read": {
"allowed_roles": ["*"],
"allowed_paths": ["~/.openclaw/workspace/"]
}
}README
# WraithVector Governance Plugin for OpenClaw
> **A Cisco audit (Feb 2026) found that 26% of ClawHub skills contain critical vulnerabilities including data exfiltration and prompt injection.** WraithVector intercepts every tool call before execution and generates cryptographic forensic evidence of every decision.
Prevent dangerous agent actions and generate verifiable audit logs for every OpenClaw tool call before execution.
**Get your free API key → [app.wraithvector.com/onboarding](https://app.wraithvector.com/onboarding)**
---
## Install (1 command)
```bash
curl -fsSL https://raw.githubusercontent.com/wraithvector0/wraithvector-openclaw-/main/install.sh | bash
```
Then set your API key:
```bash
export WRAITHVECTOR_API_KEY=wv_your_key_here
openclaw gateway restart
```
Get your free key at **[app.wraithvector.com/onboarding](https://app.wraithvector.com/onboarding)** — 30 seconds, no credit card.
---
## Threats blocked
- Unauthorized shell commands (`rm -rf`, `curl | bash`, data exfiltration)
- Sensitive path access outside allowed workspace
- Prompt injection and malicious tool usage
- Agent tool misuse and scope creep
- Unauthorized actions by compromised ClawHub skills
---
## Demo
```
agent : run the command: rm -rf /
WraithVector → BLOCK
Reason: COMMAND_NOT_ALLOWED
Evidence: cryptographic hash chain record
PDF: https://app.wraithvector.com/api/audit/pdf?id=...
```
```
agent : run the command: ls
WraithVector → ALLOW
Reason: COMMAND_IN_POLICY
Evidence: cryptographic hash chain record
```
---
## What it does
Every tool call your OpenClaw agent makes is evaluated by WraithVector **before execution**:
| Hook | What it controls |
|------|-----------------|
| `exec` | Command scope — only allowed commands execute |
| `read` | Path scope — only allowed paths are accessible |
| `write` | *(coming soon)* File write restrictions |
| `web_fetch` | *(coming soon)* Domain allowlist/blocklist |
Every ALLOW and BLOCK decision generates a **cryptographic audit trail** — independently verifiable, EU AI Act and DORA compliant.
---
## Why this matters
In February 2026, an OpenClaw agent autonomously created a MoltMatch dating profile without explicit user consent. WraithVector would have:
- **Blocked** the action before execution
- **Generated forensic evidence** of the attempt with a cryptographic hash chain
- **Alerted** the operator via Slack/email in real time
This is not hypothetical. OpenClaw agents execute real actions — shell commands, file reads, web requests — on your machine. Without governance, you have no visibility and no control.
---
## Architecture
```
OpenClaw agent tool call
↓
before_tool_call hook
↓
WraithVector API (policy evaluation)
↓
ALLOW / BLOCK
↓
+ cryptographic audit record
+ PDF evidence pack
+ real-time alert (optional)
↓
OpenClaw executes or stops
```
**Fail-closed by default** — if WraithVector is unreachable, actions are blocked to protect your system.
Set `WRAITHVECTOR_FAIL_OPEN=true` to allow actions when offline (development only).
---
## Default policy
Out of the box, WraithVector ships with a safe default policy:
```json
{
"exec": {
"allowed_roles": ["*"],
"allowed_commands": ["ls", "pwd", "echo", "cat", "grep"]
},
"read": {
"allowed_roles": ["*"],
"allowed_paths": ["~/.openclaw/workspace/"]
}
}
```
Customize your policy from your dashboard at `app.wraithvector.com`.
---
## What you get in the audit trail
Every decision generates:
- **Decision record** — tool name, command, decision (ALLOW/BLOCK), reason, timestamp
- **Cryptographic hash chain** — tamper-evident, independently verifiable
- **PDF evidence pack** — ready for compliance audits (EU AI Act Article 12, DORA)
- **Real-time alerts** — Slack and email on high-risk events
---
## Manual install (alternative)
```bash
mkdir -p ~/.openclaw/workspace/plugins/wraithvector
curl -o ~/.openclaw/workspace/plugins/wraithvector/index.mjs \
https://raw.githubusercontent.com/wraithvector0/wraithvector-openclaw-/main/index.mjs
curl -o ~/.openclaw/workspace/plugins/wraithvector/openclaw.plugin.json \
https://raw.githubusercontent.com/wraithvector0/wraithvector-openclaw-/main/openclaw.plugin.json
openclaw gateway restart
```
---
## Roadmap
- [x] `exec` tool policy enforcement
- [x] `read` tool policy enforcement
- [x] Cryptographic audit trail (hash chain)
- [x] PDF evidence pack generation
- [x] Real-time Slack/email alerts
- [ ] `write` tool policy
- [ ] `web_fetch` domain restrictions
- [ ] `after_tool_call` result governance
- [ ] Human-in-the-loop via `/approve` (OpenClaw async hooks)
- [ ] Multi-agent session governance
---
## Contributing
Contributions welcome. Open issues:
- Add `write` tool policy
- Add `web_fetch` domain allowlist/blocklist
- Human-in-the-loop integration via OpenClaw async hooks (`/approve`)
- `after_tool_call` result filtering
---
## License
MIT
---
*WraithVector — AI Agent Governance for regulated environments.*
*EU AI Act · DORA · GDPR*
Questions: [email protected]
tools
Comments
Sign in to leave a comment