A robust constitutional enforcement engine designed to govern AI agent tool calls through a strict 11-article security framework.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install cord-sentinel
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install cord-sentinel using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Cord Sentinel?
Cord Sentinel serves as the Counter-Operations & Risk Detection layer for AI agents, providing a mandatory pre-flight enforcement mechanism for high-risk operations. By integrating this engine into your Openclaw Skills, you ensure that every tool call—including shell execution, file writing, and network requests—is evaluated against a comprehensive 11-article constitution. This system prioritizes security, ethics, and operational integrity, preventing unauthorized actions before they occur.
The engine goes beyond simple filtering by scanning external data for prompt injection and providing a structured verdict system. Whether you are building complex automations or simple scripts, Cord Sentinel acts as a guardian, evaluating the second-order consequences of agent actions. It is an essential component for developers using Openclaw Skills who require a higher level of trust and safety in their autonomous systems.
Cord Sentinel Use Cases
- Preventing unauthorized shell command execution via the exec tool.
- Scanning incoming emails and web content for malicious prompt injection before processing.
- Enforcing ethical and moral constraints on AI-generated messages and financial transactions.
- Auditing agent drift and maintaining chain integrity during long-running work sessions.
- Implementing intent locks to restrict agent access to specific file paths and network domains.
How Cord Sentinel Works
- The agent identifies a high-risk tool call (e.g., exec, write, or network) or receives external data.
- A Proposal object is generated containing the tool name, input text, and required grants.
- The Proposal is sent to the CORD engine bridge for evaluation against the 11 constitutional articles.
- The engine calculates a risk score and checks for mandatory hard blocks like security or morality violations.
- A verdict is returned (ALLOW, CONTAIN, CHALLENGE, or BLOCK) with a detailed list of reasons and violated articles.
- The agent executes the action only if permitted, or reports the specific violations to the user if blocked.
Cord Sentinel Setup
Install Python 3 and ensure the bridge script is accessible. You can call the evaluation engine via the command line or import it directly into your Python-based Openclaw Skills.
# Example CLI pre-flight check
echo '{"text":"ls -la","tool_name":"exec","action_type":"command"}' | python3 -m cord_engine.bridge
To integrate via Python:
import sys
sys.path.insert(0, "~/ClaudeWork/artificial-persistent-intelligence")
from cord_engine import evaluate, Proposal
# Define the action for Openclaw Skills
proposal = Proposal(text="git commit -m 'update'", tool_name="exec", grants=["shell"])
verdict = evaluate(proposal)
print(verdict.decision)
Cord Sentinel Data Schema & Taxonomy
Cord Sentinel organizes data through a structured proposal and verdict system to ensure transparency across all Openclaw Skills.
| Field | Description | Type |
|---|---|---|
| tool_name | The specific tool being requested (exec, write, etc.) | String |
| text | The actual command or content to be processed | String |
| score | Numerical risk value from 0.0 to 10.0 | Float |
| violations | List of constitutional articles breached | Array |
| intent_lock | Current session scope and allowed path constraints | Object |
Cord Sentinel Advanced Features
- Intent Locking: Start sessions by defining explicit scopes for paths, commands, and network targets within Openclaw Skills.
- Hard Block Bypass: Instant blocking of critical security threats such as Article VII violations without waiting for a score calculation.
- Audit Log Integration: Maintain a tamper-proof record of all allowed and blocked actions for post-incident analysis.
- Multi-Source Scanning: Specifically designed to handle raw input from external sources like webhooks or user-provided files.
- Proactive Reasoning Gate: A six-question evaluation framework for high-significance actions to predict long-term consequences.
SKILL.md
Loading
Related Openclaw Skills
Labradoc CLI Skill
A powerful interface for managing Labradoc document intelligence, tasks, and file operations directly via the CLI.
Bring! Shopping List
A high-performance AI agent skill for managing Bring! shopping lists through natural language commands and automated synchronization.
Google Docs API Integration
A developer-focused skill for programmatically creating, formatting, and managing Google Documents via the official REST API.
Agent Email CLI
A robust command-line interface designed for AI agents to create disposable inboxes, monitor incoming mail, and extract message content programmatically.
DingTalk API Skill
A comprehensive integration for the DingTalk Open Platform that enables automated messaging, organizational management, and real-time event processing.
OpenQQ Bot
A high-performance QQ bot integration for OpenClaw that enables AI-driven automated replies in private and group conversations.
Featured*
