A mechanical enforcement suite that replaces soft markdown suggestions with hard git hooks and automated validation scripts for AI agents.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install jz-agent-guardrails
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 jz-agent-guardrails 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 Agent Guardrails?
Agent Guardrails is a technical framework designed to solve the common problem of AI agents ignoring project standards as context length increases. While markdown-based rules are often treated as suggestions by LLMs, this skill implements a hierarchy of mechanical enforcement that turns those rules into laws. It is specifically built for developers using Openclaw Skills who require high-reliability code generation and strict architectural adherence.
By moving enforcement from prompts to code hooks, the skill prevents expensive production incidents such as server crashes, token leaks, and redundant code rewrites. It provides a robust safety net for tools like Claude Code, Clawdbot, and Cursor, ensuring that every agent-driven change is validated against project-specific constraints and security protocols.
Agent Guardrails Use Cases
- Preventing AI agents from hardcoding sensitive secrets, tokens, or passwords into the source code.
- Stopping agents from reimplementing existing logic by enforcing module registry checks before file creation.
- Automating deployment verification to ensure that new code is correctly integrated into production pipelines.
- Implementing a meta-enforcement loop that updates the core skill whenever a project-specific guardrail is improved.
How Agent Guardrails Works
- The user initializes the skill which installs git pre-commit hooks and copies validation scaffolding into the project.
- Before an agent creates a new file, the pre-create check script analyzes existing modules to prevent code duplication.
- After the agent generates code, a post-creation validation script detects bypass patterns, missing imports, or structural errors.
- During the commit process, mechanical hooks scan for hardcoded secrets and verify that deployment checklists are satisfied.
- A feedback loop detects improvements made to project guardrails and prompts the user to sync those changes back to the global skill repository.
Agent Guardrails Setup
To install the guardrails into your current project, run the installation script provided in the package:
cd your-project/
bash /path/to/agent-guardrails/scripts/install.sh
To enable deployment verification, run the following command to generate the necessary templates:
bash scripts/create-deployment-check.sh /path/to/project
Agent Guardrails Data Schema & Taxonomy
The skill organizes its enforcement data through a combination of scripts, hooks, and templates to ensure Openclaw Skills operate within defined boundaries:
| Component | Purpose | Location |
|---|---|---|
| git-hooks | Mechanical blocks for secrets and bypass patterns | .git/hooks/ |
| Registry Templates | Defines the architecture for module imports | registry-template.py |
| Deployment Check | Scripted verification of production readiness | .deployment-check.sh |
| AGENTS.md | Project-specific rules linked to mechanical hooks | AGENTS.md |
Agent Guardrails Advanced Features
- Multi-agent compatibility with Claude Code, Clawdbot, and Cursor workflows.
- Skill Update Feedback Loop that enables automated detection and semi-automatic commits for guardrail improvements.
- Hierarchy-based enforcement that prioritizes code-level hooks (100% reliability) over markdown rules (40% reliability).
- Automated secret detection pre-integrated into the git workflow to prevent credential leaks.
SKILL.md
Loading
Related Openclaw Skills
context-doctor
Visualize and diagnose context window allocation and token usage for AI coding agents.
AWI (Agentic Web Interface)
AWI is a zero-configuration single-binary tool for AI agents to read webpages and search the internet using an intelligent three-level fallback strategy.
Mission Control
A macOS-native web command center for monitoring agent activity, managing cron jobs, and tracking LLM costs in real-time.
Plane CLI
A robust command-line interface for Plane.so that enables developers and AI agents to manage projects, issues, and states using structured JSON output.
Humanizer
A writing editor skill designed to identify and eliminate the telltale signs of machine-generated text to restore natural human rhythm and voice.
Roundtable
A sophisticated prompting framework that simulates a multi-expert panel to provide diverse viewpoints and consensus for complex decision-making.
Featured*
