A runtime-level security plugin for OpenClaw that prevents AI agents from executing development tasks without completing a mandatory design review.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install grill-gate
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 grill-gate 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 Grill Gate?
Grill Gate is an advanced runtime-level enforcement plugin designed to resolve the critical issue of AI agents bypassing design guidelines. While prompt-level constraints (like instructing an agent to 'always design before coding') are easily ignored or bypassed by raw execution commands, Grill Gate injects control directly into the execution environment. By hooking into the core runtime engine, it acts as an unbypassable gatekeeper for tool calls.
As part of the expanding ecosystem of Openclaw Skills, this plugin intercepts high-risk activities like spawning subagents or running code execution tools. It ensures that agents strictly adhere to a structured development cycle, forcing them to think, plan, and obtain a valid design token before executing tasks.
Grill Gate Use Cases
- Enforcing Agent Guardrails: Preventing autonomous agents from writing and executing code before a human or an automated design review process is completed.
- Preventing Prompt Bypasses: Blocking agents from ignoring instructions in AGENTS.md by executing raw commands via terminal tools.
- Multi-Agent Coordination: Ensuring that subagents spawned during a complex workflow are explicitly authorized through secure token validation.
- Automated Quality Gates: Integrating design review verification directly into continuous integration and agent dispatch workflows.
How Grill Gate Works
- Interception: The AI agent attempts to run a blocked command (such as launching a coding agent CLI) or spawn a sub-session.
- Hook Trigger: The plugin intercepts this attempt at the runtime layer using the unbypassable OpenClaw before_tool_call hook.
- Content Scan: Grill Gate parses the request and evaluates the command and arguments against configured trigger keywords (e.g., research, design, refactor).
- Token Verification: If a trigger or blocked command is detected, the plugin checks the designated directory for a valid, unexpired grill token.
- Execution Control: If a valid token is found, the call is permitted and the token is consumed. If no token exists, the tool call is rejected immediately, blocking execution.
Grill Gate Setup
To configure the plugin, create a configuration file at ~/.openclaw/grill-gate.json (all fields optional):
{
"triggers": ["research", "develop", "design", "architect", "plan", "refactor", "migrate"],
"exemptions": ["ASAP", "skip grill", "just do it"],
"blockedCommands": ["hermes", "claude", "codex"],
"tokenDir": "/path/to/.grill-tokens",
"tokenTtlSeconds": 3600
}
Or set via environment variable:
export GRILL_GATE_CONFIG='{"triggers":["research","design"],"tokenTtlSeconds":7200}'
To issue a new grill token after completing a design review session:
python3 scripts/auto_dispatch.py --issue-grill-token "task description"
Tokens are one-hour, one-use files stored in the configured tokenDir.
Grill Gate Data Schema & Taxonomy
Grill Gate organizes execution settings, keyword presets, and validation tokens structurally.
Configuration Options
| Parameter | Type | Description | Default |
|---|---|---|---|
triggers |
Array | Keywords in agent tasks that trigger the token check. | Built-in presets |
exemptions |
Array | Phrase patterns that bypass the token check. | Built-in presets |
blockedCommands |
Array | Executables and shell commands that require a token. | [] |
tokenDir |
String | Local directory path where active token files reside. | ~/.grill-tokens |
tokenTtlSeconds |
Integer | Lifespan of an issued token in seconds. | 3600 |
Language Presets
Built-in presets for Chinese and English keywords are both loaded by default:
| Language | Triggers | Exemptions |
|---|---|---|
| Chinese | 研究, 开发, 设计, 架构, 方案, 规划, 重构, 迁移 | 快点, 直接做, 赶紧 |
| English | research, develop, design, architect, plan, refactor, migrate | ASAP, skip grill, just do it |
Token File Schema
Tokens are stored in the token directory as individual JSON files containing validation parameters:
{
"token_id": "abc123xyz",
"status": "issued",
"expires_at": 1718900000,
"task_context": "Refactor database migration script"
}
Grill Gate Advanced Features
- Runtime Interception Hook: Integrates with the
before_tool_callhook directly at the OpenClaw execution engine tier, presenting a bypass-proof gate. - Three-Layer Security Architecture: Acts as the ultimate catch-all enforcement layer behind
auto_dispatch.pyandhermes_exec.py. - Exemption Engine: Allows specific phrases like 'ASAP' or 'skip grill' to emergency bypass checking when necessary.
- Bilingual Keyword Support: Comes with pre-configured developer vocabularies in both English and Chinese.
- Auto-Expiring Session Tokens: Implements automatic TTL expiration for security hygiene, preventing token reuse after its active session windows.
SKILL.md
Loading
Related Openclaw Skills
Pre-flight Dashboard Audit
A universal React dashboard audit skill designed to detect hardcoded values, tier bypasses, stale components, and onboarding inconsistencies before production release.
Richard Nixon Biography & Leadership Toolkit
An executable strategic toolkit for AI agents to extract leadership, resilience, and cautionary lessons from the legendary rise and fall of Richard Nixon.
Task Planner
A personal task and schedule management assistant that interprets natural language to create, edit, recur, and export schedules and reminders.
Ten Women Who Changed Science and the World
An executable toolkit that leverages AI agents to extract powerful lessons, scientific discoveries, and resilience strategies from ten pioneering female scientists.
Power, Sex, Suicide: Mitochondrial Biology Toolkit
An executable knowledge agent that explains how mitochondria shape life, evolution, sex, aging, and cell death based on Nick Lane's pioneering book.
Radically Happy Mind Training Skill
An executable mental training skill that bridges Buddhist wisdom with modern psychology to help users cultivate mindfulness, dissolve negative habits, and build lasting happiness.
Featured*
